diff --git a/.gitattributes b/.gitattributes index bed0738c7eeb449bca98b5d2f33c89a1ee56349a..9bc10dfd21aeb6b09817d77bb8962e47eb03fec4 100644 --- a/.gitattributes +++ b/.gitattributes @@ -1,60 +1,584 @@ *.7z filter=lfs diff=lfs merge=lfs -text *.arrow filter=lfs diff=lfs merge=lfs -text -*.avro filter=lfs diff=lfs merge=lfs -text *.bin filter=lfs diff=lfs merge=lfs -text +*.bin.* filter=lfs diff=lfs merge=lfs -text *.bz2 filter=lfs diff=lfs merge=lfs -text -*.ckpt filter=lfs diff=lfs merge=lfs -text *.ftz filter=lfs diff=lfs merge=lfs -text *.gz filter=lfs diff=lfs merge=lfs -text *.h5 filter=lfs diff=lfs merge=lfs -text *.joblib filter=lfs diff=lfs merge=lfs -text *.lfs.* filter=lfs diff=lfs merge=lfs -text -*.lz4 filter=lfs diff=lfs merge=lfs -text -*.mds filter=lfs diff=lfs merge=lfs -text -*.mlmodel filter=lfs diff=lfs merge=lfs -text *.model filter=lfs diff=lfs merge=lfs -text *.msgpack filter=lfs diff=lfs merge=lfs -text -*.npy filter=lfs diff=lfs merge=lfs -text -*.npz filter=lfs diff=lfs merge=lfs -text *.onnx filter=lfs diff=lfs merge=lfs -text *.ot filter=lfs diff=lfs merge=lfs -text *.parquet filter=lfs diff=lfs merge=lfs -text *.pb filter=lfs diff=lfs merge=lfs -text -*.pickle filter=lfs diff=lfs merge=lfs -text -*.pkl filter=lfs diff=lfs merge=lfs -text -*.pt filter=lfs diff=lfs merge=lfs -text + *.pth filter=lfs diff=lfs merge=lfs -text *.rar filter=lfs diff=lfs merge=lfs -text -*.safetensors filter=lfs diff=lfs merge=lfs -text saved_model/**/* filter=lfs diff=lfs merge=lfs -text -*.tar.* filter=lfs diff=lfs merge=lfs -text *.tar filter=lfs diff=lfs merge=lfs -text +*.tar.* filter=lfs diff=lfs merge=lfs -text +*.mat filter=lfs diff=lfs merge=lfs -text +*.npz filter=lfs diff=lfs merge=lfs -text +*.npy filter=lfs diff=lfs merge=lfs -text +*.h5 filter=lfs diff=lfs merge=lfs -text +*.hdf5 filter=lfs diff=lfs merge=lfs -text +*.pickle filter=lfs diff=lfs merge=lfs -text +*.pkl filter=lfs diff=lfs merge=lfs -text *.tflite filter=lfs diff=lfs merge=lfs -text *.tgz filter=lfs diff=lfs merge=lfs -text -*.wasm filter=lfs diff=lfs merge=lfs -text *.xz filter=lfs diff=lfs merge=lfs -text *.zip filter=lfs diff=lfs merge=lfs -text -*.zst filter=lfs diff=lfs merge=lfs -text -*tfevents* filter=lfs diff=lfs merge=lfs -text -# Audio files - uncompressed -*.pcm filter=lfs diff=lfs merge=lfs -text -*.sam filter=lfs diff=lfs merge=lfs -text -*.raw filter=lfs diff=lfs merge=lfs -text -# Audio files - compressed -*.aac filter=lfs diff=lfs merge=lfs -text -*.flac filter=lfs diff=lfs merge=lfs -text -*.mp3 filter=lfs diff=lfs merge=lfs -text -*.ogg filter=lfs diff=lfs merge=lfs -text -*.wav filter=lfs diff=lfs merge=lfs -text -# Image files - uncompressed -*.bmp filter=lfs diff=lfs merge=lfs -text -*.gif filter=lfs diff=lfs merge=lfs -text -*.png filter=lfs diff=lfs merge=lfs -text -*.tiff filter=lfs diff=lfs merge=lfs -text -# Image files - compressed +*.zstandard filter=lfs diff=lfs merge=lfs -text +*.tfevents* filter=lfs diff=lfs merge=lfs -text +*.db* filter=lfs diff=lfs merge=lfs -text +*.ark* filter=lfs diff=lfs merge=lfs -text +**/*ckpt*data* filter=lfs diff=lfs merge=lfs -text +**/*ckpt*.meta filter=lfs diff=lfs merge=lfs -text +**/*ckpt*.index filter=lfs diff=lfs merge=lfs -text +*.safetensors filter=lfs diff=lfs merge=lfs -text +*.ckpt filter=lfs diff=lfs merge=lfs -text *.jpg filter=lfs diff=lfs merge=lfs -text +*.png filter=lfs diff=lfs merge=lfs -text *.jpeg filter=lfs diff=lfs merge=lfs -text +*.bmp filter=lfs diff=lfs merge=lfs -text +*.gif filter=lfs diff=lfs merge=lfs -text *.webp filter=lfs diff=lfs merge=lfs -text -# Video files - compressed +*.mp3 filter=lfs diff=lfs merge=lfs -text +*.wav filter=lfs diff=lfs merge=lfs -text +*.wma filter=lfs diff=lfs merge=lfs -text +*.aac filter=lfs diff=lfs merge=lfs -text +*.ogg filter=lfs diff=lfs merge=lfs -text +*.m4a filter=lfs diff=lfs merge=lfs -text +*.m3u8 filter=lfs diff=lfs merge=lfs -text +*.amr filter=lfs diff=lfs merge=lfs -text +*.audio filter=lfs diff=lfs merge=lfs -text +*.avi filter=lfs diff=lfs merge=lfs -text +*.flv filter=lfs diff=lfs merge=lfs -text *.mp4 filter=lfs diff=lfs merge=lfs -text -*.webm filter=lfs diff=lfs merge=lfs -text +*.mpg filter=lfs diff=lfs merge=lfs -text +*.asf filter=lfs diff=lfs merge=lfs -text +*.mov filter=lfs diff=lfs merge=lfs -text +*.mpeg filter=lfs diff=lfs merge=lfs -text +*.3gp filter=lfs diff=lfs merge=lfs -text +*.wmv filter=lfs diff=lfs merge=lfs -text +*.rmvb filter=lfs diff=lfs merge=lfs -text +*.rm filter=lfs diff=lfs merge=lfs -text +*.ts filter=lfs diff=lfs merge=lfs -text +*.mkv filter=lfs diff=lfs merge=lfs -text +*.flash filter=lfs diff=lfs merge=lfs -text +*.vob filter=lfs diff=lfs merge=lfs -text +*.pdf filter=lfs diff=lfs merge=lfs -text +*.ost filter=lfs diff=lfs merge=lfs -text +*.pst filter=lfs diff=lfs merge=lfs -text +*.doc filter=lfs diff=lfs merge=lfs -text +*.docx filter=lfs diff=lfs merge=lfs -text +*.txt filter=lfs diff=lfs merge=lfs -text +*.ppt filter=lfs diff=lfs merge=lfs -text +*.pptx filter=lfs diff=lfs merge=lfs -text +*.xls filter=lfs diff=lfs merge=lfs -text +*.xlsx filter=lfs diff=lfs merge=lfs -text +*.vsd filter=lfs diff=lfs merge=lfs -text +*.vsdx filter=lfs diff=lfs merge=lfs -text +*.jsonl filter=lfs diff=lfs merge=lfs -text +*.json filter=lfs diff=lfs merge=lfs -text +dataset_infos.json ignore +*.csv filter=lfs diff=lfs merge=lfs -text +*.tsv filter=lfs diff=lfs merge=lfs -text + +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00333-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00524-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00328-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00512-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00731-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01095-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00175-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00657-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00332-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00282-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00933-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00186-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01347-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00927-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00688-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00693-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00153-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01135-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01413-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00334-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00384-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00350-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00127-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00358-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00126-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01421-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01390-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00001-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00385-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00557-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00360-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00559-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00758-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00136-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00196-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00131-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00132-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00570-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00193-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00167-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01161-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01125-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00538-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00764-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01162-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00845-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00086-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00085-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00090-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00325-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00316-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00272-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00116-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01195-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00461-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00324-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00471-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00445-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00464-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00314-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00644-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00462-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00472-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00643-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00475-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00476-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00327-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00465-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00657-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00512-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00272-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00278-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00688-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00676-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00731-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00538-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00693-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01201-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00186-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01095-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00524-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00061-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00927-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00282-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00175-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00061-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01162-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01135-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00090-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01125-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01161-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00086-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00085-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00116-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01195-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00083-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00325-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01163-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00351-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01441-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00084-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00324-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00314-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00327-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01422-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01347-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00128-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00316-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00328-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00332-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00333-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00870-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00334-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01413-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01422-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01421-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01390-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00361-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00360-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01441-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00351-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00127-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00557-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00350-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00358-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00384-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00128-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00136-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00385-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00193-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00001-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00132-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01163-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00126-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00196-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00570-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00131-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00845-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00153-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00559-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01296-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00870-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00084-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00758-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01313-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00167-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01296-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00208-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01201-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00764-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00620-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00637-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00042-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00032-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00041-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00410-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00412-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00636-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00444-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00461-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00411-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00413-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00443-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00462-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00446-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00433-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00471-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00475-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00512-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00445-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00657-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00447-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00465-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00463-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00643-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00472-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00464-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00175-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00476-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00061-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00272-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00933-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00644-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00282-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00278-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00538-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00676-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00731-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00693-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00688-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01135-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01195-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00083-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00927-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00186-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01125-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01163-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01161-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00333-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00084-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00324-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00116-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00327-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00524-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00085-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00086-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01095-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01313-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01162-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00314-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00325-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00332-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00328-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01347-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01390-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00316-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01441-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00358-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00384-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00090-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00128-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00126-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00127-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01421-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00360-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00350-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00385-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00193-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01413-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00334-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00570-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01422-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00136-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00132-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00196-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00351-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00131-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00557-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00208-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00041-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00001-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00167-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00361-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00758-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00153-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00559-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00411-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00764-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00208-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00845-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00042-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00032-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00620-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00411-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00042-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00870-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00412-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01296-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00433-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00032-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01313-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01201-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00636-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00041-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00636-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00637-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00620-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00637-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00461-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00446-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00447-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00413-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00410-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00412-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00443-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00444-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00446-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00116-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00462-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00463-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00472-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00445-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00464-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00465-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00471-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00445-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00644-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00433-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00475-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00278-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00413-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00316-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00444-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00461-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00443-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00314-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00476-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00463-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00643-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00328-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00462-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00447-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00471-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00334-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00464-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00332-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00325-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00333-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00465-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00643-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00472-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01347-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00644-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01390-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00512-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00327-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00476-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00278-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00475-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01422-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01421-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00688-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00272-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01441-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01413-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00676-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00351-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00061-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00358-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00693-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01201-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00557-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00350-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00360-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00361-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00001-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00385-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00126-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00186-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00731-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00657-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00128-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00384-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00132-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00127-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00196-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00463-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00193-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00131-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00136-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00570-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00559-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00175-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00524-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00167-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00282-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00764-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00927-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00538-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00032-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00845-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00153-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00870-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01161-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00041-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00620-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00272-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01296-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00208-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00090-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01095-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01125-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00042-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01162-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00933-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00636-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00676-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01313-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00657-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00512-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00688-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00410-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00413-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00637-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00433-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00411-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00412-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00175-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01135-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00443-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01163-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00446-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00731-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00445-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00084-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00444-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00693-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00090-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00086-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00083-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00461-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00282-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00085-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01195-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00085-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00086-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00116-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00084-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00324-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00316-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00472-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00471-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00314-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01422-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00462-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00538-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00447-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00465-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00061-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00463-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00328-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00333-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00643-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00327-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00325-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01347-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00475-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00524-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00644-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01125-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00186-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00933-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00278-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01163-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01161-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00927-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01421-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00083-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01095-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00476-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01195-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00358-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00334-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00332-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01135-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01390-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01413-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01162-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00351-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00153-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00360-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01441-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00361-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00758-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00384-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00131-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00128-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00350-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00126-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00557-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00127-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00042-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00385-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01313-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00041-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00001-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00208-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00132-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00764-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00410-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00167-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00411-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00413-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00559-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00032-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01201-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00433-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00196-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00570-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00870-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00933-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00636-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00447-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00136-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00444-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00620-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01296-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00193-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00637-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00845-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00443-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00412-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00446-wflip.zst filter=lfs diff=lfs merge=lfs -textlambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00676-wflip.zst filter=lfs diff=lfs merge=lfs -text +lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00083-wflip.zst filter=lfs diff=lfs merge=lfs -text diff --git a/README.md b/README.md new file mode 100644 index 0000000000000000000000000000000000000000..c69d0b2a93dac6d254eb984955169caa229f2cef --- /dev/null +++ b/README.md @@ -0,0 +1,9 @@ +--- +license: mit +tags: + - large-model-feature-coding +models: + - facebook/dinov3-vit7b16-pretrain-lvd1689m +language: + - en +--- diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00512-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00512-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..a215e1149e9b9628ed499ef170546e5d594a4af3 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00512-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e5b42e4f66b924e4094392341b00e657c6692450ad8180c50234cf728618be59 +size 90811944 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00657-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00657-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..52a9bae6a681df256b917b099df462f883c6f161 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00657-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:106666e562fe18022d6ffbbd59f9bef8953381ad558f3a0e38b89181297e0c93 +size 88782429 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00676-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00676-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..55be428ca19d13b2c0e9639fac2521b84515ee6e --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00676-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d286ad71b4f7eab7ec33b5b7be69a827e9987895d3828094b52c2d8fb0027517 +size 91132469 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00688-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00688-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..4a1bf00154494f4fe1c1cf873b3de87b5a4b4d39 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00688-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:ad1802c2b89e71e21c7ddab2adde7550cb32dd43b6f018432d37cd9c1f6ed43d +size 89193475 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00693-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00693-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..ae8a47d3af761a1567bb123d6807fa273d6c39c0 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00693-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:447441a7994423d311cd2c8b51bc9fa83a38ddf8bac3bcec96d79710fbeba075 +size 91480198 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00731-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00731-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..dbf14ebea44be900c4ab183f7302179db7b61347 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00731-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d2a16b767ed280c793099fe5874da89ffe20108337c1160667cbb8e39aebc0d9 +size 90951474 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00061-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00061-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..c3af101a872c1041015f9ffe56494bb4b3943c30 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00061-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:848b077d7c9d6113ecb37281fe5987ba36ab7e78803e9a6ab4e748c4b97749d8 +size 90348496 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00175-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00175-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..eaec905e4ec3773f3dd9e609fd5c884c6f3fe195 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00175-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3f6daff01ad6b8a388947b989aafd3547b759dff874ed16d755d66cbe37344f8 +size 89547760 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00186-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00186-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..45d365059a12fc9f74c90a72864e917fb2e020f1 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00186-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a1ce015d86e2df1687590a289200c9ef2b651939e79d735a9d735244baa968e0 +size 89171371 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00282-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00282-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..906c6db8a4e3a4ac806f6eb1bdeaeb5b778bc3a9 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00282-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:58ae51e1d7328c6471a6f54a3f8b274eb5c3d7b57b8dd7f3f8c55f9544e306fb +size 90958725 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00524-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00524-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..ed0f5d0dcd44feddf8d3e8c2b13756e96f51eb03 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00524-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:49b8dad8c80b5b2a40f1e74b87dc6a8ac093a4c17f9d215f71231b57736ebcf8 +size 91573345 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00538-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00538-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..ac633e9cce1cc814884e54a8ea1e36626317ff7d --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00538-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bc45cf2855a0a73344f9d0cef83d56dbb9d381f7feb3c094672dcb05912ac39f +size 90255634 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00927-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00927-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..bcab9b83b673fcab61e260c2b35d617dfa712d28 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00927-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05b02dcf3a53e349095fbb627d29ab6f64ea7264b52b4405bc80b85d057090f6 +size 91754145 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00933-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00933-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..b53910f8db466be438c0d66178e6c606c62a53b4 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00933-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8e3fb3593082e271495be8215f29b3e8107098f4afec7c70affed55cfe8562ac +size 91185468 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01095-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01095-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..90e88fc502457906b1287fc872b03f99cd2a3699 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01095-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:56b3fafa8c84f87148012c1e035013f3a50bea8e6ed0033f2e1e67d05cfd917b +size 91048927 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01125-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01125-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..de959974f3e2e169a836efbd2eb556510285264c --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01125-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f80d5b9e363bd20e2ee08401e11859866d027b5c21886b4570d1920a654cd471 +size 90415306 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01135-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01135-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..1fc2383bde98e376c97d137d4c45c0be8b9f0935 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01135-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d20581be8bdc6d34ce75cf98d3913627fafe94172d9da86a593bf21a24ec87b5 +size 89950505 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01161-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01161-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..22e56d9950ad6d69aaed13b47b5747d67ec0fc19 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01161-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a82ddbe26bb959f4808133c554e37df04f1bee8f61842eb52acddd444928dc6f +size 89652054 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01162-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01162-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..28d665e10716fc1bc9d5351fc6119a82048cf1dd --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01162-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:6a840f75033e1795be5ed74e08a8d01fa0b1f942a3f17294e31f494547e377ac +size 90952504 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01163-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01163-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..d554b6808262c95cf1ec39420f47356795caaef0 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01163-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3d0630d9d58bbcc0375a410d4140a123a8c68559fda78230845f0d3bc2516beb +size 90247056 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01195-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01195-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..752b2350e3f8891040ffdcf5f42ad2aeb557b9bc --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01195-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:2648602232d013dcfa4f4c73e9c1a450742a404112d42ed1be7e3db70e65fbab +size 91028812 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00083-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00083-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..fee4ee26e5fe3159d4190e208725c1d5f88e8ca6 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00083-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:213ff9ee236f6fc8a8eac982134e99c2f20eb89e649062e56ec39e452b396cc8 +size 90886099 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00084-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00084-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..40f4ac6566aa2ee2156a78347d098a06eb9d0cb2 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00084-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:83e33c1d896eec0c16a8ab87c219cb21b5808dfef7071e7b0d7b719c1b7982d3 +size 91986047 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00085-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00085-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..5e43bb4e6708d56faa9a29dba470cfb197050b61 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00085-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:741e0ea254c0781b18af1921ffb9f08eb1765ca49bfe5b99f26e1ab6be463547 +size 88304089 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00086-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00086-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..7b9aaded40f128c3baac808914fc8703e5bd9965 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00086-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:40898d447c6a88b17e2618c420592c280468ef123585ce6f6ce96806c0de4ca7 +size 88469474 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00090-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00090-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..00022a872ca21a6fbacd1116994f688c62dcd2db --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00090-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:20763b3493348c54927558c8bcfcf221e9cdf7580bcde8b233ef1bdc17d419e3 +size 90079170 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00116-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00116-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..ebbba92675dd9d2a51e1eae54390d711a456630e --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00116-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8946c928885767a8f1c21d90390532001d99afa52f3f60b0636a5d6c195a1f9d +size 88954622 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00314-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00314-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..d070e862d5b2415ec50a9dbb641a47208fedf3e9 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00314-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d5fd50402bea4b5a31f3c58c9ec75109d3f9bf75b605a2b8c46947573312ccbd +size 90401595 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00316-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00316-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..550b1482b6e657fd6b912472007de5874b480cbb --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00316-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:cec3c96dc937ca2b18f93d169086e4debbddd80594b03c870104bc9171d5af7a +size 90126857 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00324-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00324-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..e3cd0fb88e7c3464d65e83d00b31088c85428c57 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00324-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:62ac782ce3ba71399d3fd066099be4155d8766cfc1c0cd62cde734c1ba0f78e5 +size 88480030 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00325-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00325-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..2916e97663e883a230418bc2579225569ec3cdf4 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00325-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:921a4ef411018a7db201f25c81cc61292224d07546b688fad5c861a6a51f534c +size 88664896 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00327-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00327-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..b3676e4a8b93328b62468707afa833cef5eb55ec --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00327-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7c3bb9664a252f96d5083ad38a443c418a33991105a77a97d752149545c6a9fd +size 89523109 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00328-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00328-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..a8b22a7e06acb330965ae8cdcd51acb9e15effac --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00328-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:8e1b77756b0788600877ed8845df6123c0dee352474656f8127bcdee50d53366 +size 89666193 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00332-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00332-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..91c3f77f4e6acd4f5814169baa33943037a7c480 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00332-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:550cb032e9306dbf2619272bb6733f4505c3b0996376d8ad66544c3f2323021f +size 90165020 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00333-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00333-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..e012a03b27b7bd71fca2aae2ac17040db0d0fd83 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00333-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:24dc13068d4423a1ff397c16121cf8c1b675833260c3670e8ec1d2d7a9508b2d +size 89943527 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00334-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00334-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..f6be7f81aa29412a9932a39cd3b5b9f2a1450309 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00334-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:bc4b043ff8abec05d5dd2e6e55c5b7346e60bc7d110ca698bd563b9509982c7a +size 90474661 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01347-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01347-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..b16966ba5b64598715341ede50955fd8a4538df2 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01347-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f0745cfecfe9f8096457ffb98107a65aedd6a9fea604dc57140872e4ece1a16f +size 90474200 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01390-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01390-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..269358bdf07f769e7552c85a6f136922ae6427b9 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01390-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:37f3b08bacee0fd5da44f4e7f56a5cc76f1fce63fa7bf13c9459e9bb545b4589 +size 90464522 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01413-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01413-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..b2439c38cc47bc36b0e6995f2998d7e9815b2d82 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01413-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d034e8368d9e442bb61f6f6ffd1ce61a3858cc05acc9d7866f834fc7b5d5f97a +size 91027576 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01421-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01421-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..864308f41b27b8ebcd418fde0cd9b603047a49b5 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01421-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:a891424ad8619779a3fdac68309f7809cebf7f729ede30ea85c51aa747f2733a +size 90304076 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01422-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01422-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..1e8c49e4e181087b7e44faa8cf3a7b7a01818485 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01422-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3084c2c65f5f979eaf184f88aaf7788a5353d4fb70cbec9b3d0dc40d509031f7 +size 88343129 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01441-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01441-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..a25af21a313450a02dd2e52e76c0017b10e1c328 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01441-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:20a178a5e6f9391eb87c5fc03861a9916a9b4d6d37026d29b266a1d501e88c67 +size 88817337 diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log new file mode 100644 index 0000000000000000000000000000000000000000..ddd2c8fef1a9d5b2009b8d22f3857cdf8ac6bb57 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log @@ -0,0 +1,15744 @@ +Experiment: dtufc_hyperprior-featurecoding_dinov3-total_individual +Log file: output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log +DTUFCCodecConfig: + arch: hyperprior-featurecoding + handler: dinov3-total + checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.001_epochs600_lr0.0001_bs360_patch256-256_checkpoint_best.pth.tar + transform_type: kmeans + transform_mapping:featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json + bit_depth: 8 + device: cuda:0 +Loading checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.001_epochs600_lr0.0001_bs360_patch256-256_checkpoint_best.pth.tar +Checkpoint epoch: 405 +Loaded hyperprior-featurecoding (1-channel) on cuda:0 +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_9_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.9' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_9_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_19_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.19' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_19_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_29_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.29' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_29_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_39_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.39' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_39_0.json +Loaded per-key mappings: model=dinov3-total + Keys: ['layer.9', 'layer.19', 'layer.29', 'layer.39'] +---------------- ------------------------------------------------------------------------------------------------------------------------------ +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +Checkpoint codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.001_epochs600_lr0.0001_bs360_patch256-256_checkpoint_best.pth.tar +Transform type kmeans +Transform config featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json +Input ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features +Output output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test +---------------- ------------------------------------------------------------------------------------------------------------------------------ +Files found: 100 +---------------------------------------------------------------------- + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst (1/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.256s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,600B, BPFP=0.0385 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,160B, BPFP=0.0382 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 80,108B, BPFP=0.0508 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,584B, BPFP=0.0499 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 98,324B, BPFP=0.0624 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 97,172B, BPFP=0.0617 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 99,260B, BPFP=0.0630 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 99,200B, BPFP=0.0630 +⌛️ [2/4] FRONTEND: Frontend time: 7.960s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.764s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14561056 5.86058610 + layer.9.1 0.14522085 5.84738458 + layer.19.0 3.25142184 24.65399791 + layer.19.1 3.25206135 24.73390275 + layer.29.0 4.23946030 275.68374228 + layer.29.1 4.24539299 278.26446214 + layer.39.0 32.17105490 20574.78843029 + layer.39.1 19.15684032 20967.02502437 + ------------------------------------------------------------------------------------- + TOTAL 8.32588289 5269.60719130 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 673408 +BPFP 0.0534 bits/point +EBPFP 0.0534 equivalent bits/point +MSE 5269.607191 +---------------------- --------------------------------------------------------- +Time: 21.981s Load: 1.256s, Pack+Encode: 7.960s, Decode+Unpack: 12.764s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5269.6072 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00512-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst (2/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.270s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,284B, BPFP=0.0364 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,264B, BPFP=0.0363 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,352B, BPFP=0.0485 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 76,408B, BPFP=0.0485 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 91,504B, BPFP=0.0581 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 92,456B, BPFP=0.0587 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,312B, BPFP=0.0592 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,876B, BPFP=0.0602 +⌛️ [2/4] FRONTEND: Frontend time: 7.738s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.623s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03307773 5.92482532 + layer.9.1 0.03291117 5.92117233 + layer.19.0 0.04156009 25.83663572 + layer.19.1 0.03760627 25.65810347 + layer.29.0 4.28582750 273.20056061 + layer.29.1 4.28551552 270.51100910 + layer.39.0 9.83402183 19138.71433214 + layer.39.1 9.85397836 19022.66493338 + ------------------------------------------------------------------------------------- + TOTAL 3.55056231 4846.05394651 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 639456 +BPFP 0.0507 bits/point +EBPFP 0.0507 equivalent bits/point +MSE 4846.053947 +---------------------- --------------------------------------------------------- +Time: 21.631s Load: 1.270s, Pack+Encode: 7.738s, Decode+Unpack: 12.623s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4846.0539 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00657-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst (3/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.286s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,692B, BPFP=0.0373 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,588B, BPFP=0.0372 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,020B, BPFP=0.0495 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,240B, BPFP=0.0490 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 98,856B, BPFP=0.0627 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 98,216B, BPFP=0.0623 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 102,676B, BPFP=0.0652 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 102,520B, BPFP=0.0651 +⌛️ [2/4] FRONTEND: Frontend time: 7.521s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.106s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00083900 6.04175384 + layer.9.1 0.00259629 6.01898283 + layer.19.0 0.00955961 26.23261293 + layer.19.1 0.08538111 26.06386852 + layer.29.0 0.11631418 288.71914608 + layer.29.1 0.11200302 287.26631053 + layer.39.0 14.47657393 19840.00259994 + layer.39.1 13.08093694 20219.92850179 + ------------------------------------------------------------------------------------- + TOTAL 3.48552551 5087.53422206 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 674808 +BPFP 0.0535 bits/point +EBPFP 0.0535 equivalent bits/point +MSE 5087.534222 +---------------------- --------------------------------------------------------- +Time: 20.913s Load: 1.286s, Pack+Encode: 7.521s, Decode+Unpack: 12.106s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5087.5342 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00676-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst (4/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.146s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,160B, BPFP=0.0363 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,272B, BPFP=0.0364 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,720B, BPFP=0.0487 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,620B, BPFP=0.0493 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 95,940B, BPFP=0.0609 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 97,068B, BPFP=0.0616 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 91,792B, BPFP=0.0583 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 93,756B, BPFP=0.0595 +⌛️ [2/4] FRONTEND: Frontend time: 7.796s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.608s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.70169554 5.85500729 + layer.9.1 0.03294074 5.81913060 + layer.19.0 3.25671692 25.95174378 + layer.19.1 3.25834093 25.95881997 + layer.29.0 0.10810242 299.62341566 + layer.29.1 0.10661203 296.41686302 + layer.39.0 8.95005916 17976.37439064 + layer.39.1 8.98756017 17920.88657784 + ------------------------------------------------------------------------------------- + TOTAL 3.42525349 4569.61074360 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 647328 +BPFP 0.0514 bits/point +EBPFP 0.0514 equivalent bits/point +MSE 4569.610744 +---------------------- --------------------------------------------------------- +Time: 21.550s Load: 1.146s, Pack+Encode: 7.796s, Decode+Unpack: 12.608s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4569.6107 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00688-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst (5/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.143s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,536B, BPFP=0.0378 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,020B, BPFP=0.0381 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,208B, BPFP=0.0496 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,756B, BPFP=0.0494 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 100,808B, BPFP=0.0640 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,772B, BPFP=0.0633 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 101,356B, BPFP=0.0643 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 100,120B, BPFP=0.0636 +⌛️ [2/4] FRONTEND: Frontend time: 7.608s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.507s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14527870 5.94006058 + layer.9.1 0.14521496 5.94085656 + layer.19.0 0.03964342 26.68300597 + layer.19.1 0.03956446 26.74579542 + layer.29.0 0.12258449 305.73224732 + layer.29.1 0.12735008 310.91231313 + layer.39.0 32.94776263 19884.24959376 + layer.39.1 29.25669534 19907.48391290 + ------------------------------------------------------------------------------------- + TOTAL 7.85301176 5059.21097320 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 677576 +BPFP 0.0538 bits/point +EBPFP 0.0538 equivalent bits/point +MSE 5059.210973 +---------------------- --------------------------------------------------------- +Time: 21.258s Load: 1.143s, Pack+Encode: 7.608s, Decode+Unpack: 12.507s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5059.2110 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00693-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst (6/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.154s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,008B, BPFP=0.0375 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,640B, BPFP=0.0372 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,776B, BPFP=0.0500 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,248B, BPFP=0.0497 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 99,716B, BPFP=0.0633 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 98,852B, BPFP=0.0627 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 97,500B, BPFP=0.0619 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 96,960B, BPFP=0.0615 +⌛️ [2/4] FRONTEND: Frontend time: 7.585s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.474s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66816044 5.94611990 + layer.9.1 2.66817504 5.95261910 + layer.19.0 3.22262959 25.74734929 + layer.19.1 3.22037432 25.59947544 + layer.29.0 4.30448692 303.06201251 + layer.29.1 4.31085282 304.76391371 + layer.39.0 38.33931691 18560.46148846 + layer.39.1 57.25219370 19079.85180370 + ------------------------------------------------------------------------------------- + TOTAL 14.49827372 4788.92309777 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 667700 +BPFP 0.0530 bits/point +EBPFP 0.0530 equivalent bits/point +MSE 4788.923098 +---------------------- --------------------------------------------------------- +Time: 21.213s Load: 1.154s, Pack+Encode: 7.585s, Decode+Unpack: 12.474s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4788.9231 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00731-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst (7/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.140s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,560B, BPFP=0.0372 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,272B, BPFP=0.0370 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,848B, BPFP=0.0488 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 76,972B, BPFP=0.0489 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 95,764B, BPFP=0.0608 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,812B, BPFP=0.0608 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 95,676B, BPFP=0.0607 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 95,564B, BPFP=0.0607 +⌛️ [2/4] FRONTEND: Frontend time: 7.630s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.143s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11332920 5.98859736 + layer.9.1 0.00092169 5.98557023 + layer.19.0 3.23006092 25.21056935 + layer.19.1 3.23257961 25.12908779 + layer.29.0 4.28548854 275.87024699 + layer.29.1 4.27808990 276.55191745 + layer.39.0 10.57841825 18415.15502112 + layer.39.1 20.33118703 18722.16184595 + ------------------------------------------------------------------------------------- + TOTAL 5.75625939 4719.00660703 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 653468 +BPFP 0.0518 bits/point +EBPFP 0.0518 equivalent bits/point +MSE 4719.006607 +---------------------- --------------------------------------------------------- +Time: 20.913s Load: 1.140s, Pack+Encode: 7.630s, Decode+Unpack: 12.143s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4719.0066 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00061-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst (8/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.220s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,716B, BPFP=0.0385 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 61,084B, BPFP=0.0388 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,776B, BPFP=0.0487 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,272B, BPFP=0.0490 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 93,672B, BPFP=0.0595 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 94,404B, BPFP=0.0599 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 90,536B, BPFP=0.0575 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 91,320B, BPFP=0.0580 +⌛️ [2/4] FRONTEND: Frontend time: 7.107s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 11.960s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14443954 6.03986292 + layer.9.1 0.14435121 6.01197646 + layer.19.0 0.03807715 25.79085503 + layer.19.1 0.03781311 25.60393139 + layer.29.0 0.10781899 284.47985050 + layer.29.1 0.10618912 283.03471319 + layer.39.0 9.30898666 16669.57556061 + layer.39.1 9.83625107 17231.38511537 + ------------------------------------------------------------------------------------- + TOTAL 2.46549086 4316.49023318 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 645780 +BPFP 0.0512 bits/point +EBPFP 0.0512 equivalent bits/point +MSE 4316.490233 +---------------------- --------------------------------------------------------- +Time: 20.287s Load: 1.220s, Pack+Encode: 7.107s, Decode+Unpack: 11.960s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4316.4902 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00175-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst (9/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.143s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,196B, BPFP=0.0369 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,952B, BPFP=0.0368 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,052B, BPFP=0.0483 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 75,756B, BPFP=0.0481 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 93,508B, BPFP=0.0594 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 94,144B, BPFP=0.0598 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 92,388B, BPFP=0.0586 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,900B, BPFP=0.0590 +⌛️ [2/4] FRONTEND: Frontend time: 7.730s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.776s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14543928 6.06682709 + layer.9.1 0.14562574 6.04912900 + layer.19.0 0.11552505 26.46107968 + layer.19.1 0.12052174 26.44065344 + layer.29.0 0.10841144 284.94619353 + layer.29.1 0.10845811 286.80080029 + layer.39.0 9.17501701 18169.95385115 + layer.39.1 9.20635778 17999.28371791 + ------------------------------------------------------------------------------------- + TOTAL 2.39066952 4600.75028151 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 640896 +BPFP 0.0509 bits/point +EBPFP 0.0509 equivalent bits/point +MSE 4600.750282 +---------------------- --------------------------------------------------------- +Time: 21.650s Load: 1.143s, Pack+Encode: 7.730s, Decode+Unpack: 12.776s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4600.7503 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00186-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst (10/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.220s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,780B, BPFP=0.0386 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,952B, BPFP=0.0387 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 79,028B, BPFP=0.0502 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,504B, BPFP=0.0498 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,896B, BPFP=0.0621 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 97,464B, BPFP=0.0619 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 95,052B, BPFP=0.0603 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,776B, BPFP=0.0602 +⌛️ [2/4] FRONTEND: Frontend time: 7.138s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.035s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.78226592 6.04011872 + layer.9.1 2.78427046 6.04159388 + layer.19.0 3.22580366 25.59023349 + layer.19.1 3.22969594 25.47106557 + layer.29.0 4.29525448 291.91871141 + layer.29.1 0.11349234 291.84703039 + layer.39.0 8.89338553 17247.28631784 + layer.39.1 8.88767087 17303.33441664 + ------------------------------------------------------------------------------------- + TOTAL 4.27647990 4399.69118599 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 664452 +BPFP 0.0527 bits/point +EBPFP 0.0527 equivalent bits/point +MSE 4399.691186 +---------------------- --------------------------------------------------------- +Time: 20.393s Load: 1.220s, Pack+Encode: 7.138s, Decode+Unpack: 12.035s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4399.6912 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00282-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst (11/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.145s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 62,420B, BPFP=0.0396 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 62,644B, BPFP=0.0398 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 80,168B, BPFP=0.0509 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,188B, BPFP=0.0503 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,912B, BPFP=0.0621 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 97,476B, BPFP=0.0619 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,192B, BPFP=0.0598 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,600B, BPFP=0.0600 +⌛️ [2/4] FRONTEND: Frontend time: 7.573s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.459s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14514284 5.97490136 + layer.9.1 0.14518188 5.97539075 + layer.19.0 0.04057091 26.09662669 + layer.19.1 0.04041447 25.92309880 + layer.29.0 4.25641542 280.90443208 + layer.29.1 4.26613502 278.17344410 + layer.39.0 12.58558458 17278.31524212 + layer.39.1 8.96866240 17434.56093598 + ------------------------------------------------------------------------------------- + TOTAL 3.80601344 4416.99050898 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 668600 +BPFP 0.0530 bits/point +EBPFP 0.0530 equivalent bits/point +MSE 4416.990509 +---------------------- --------------------------------------------------------- +Time: 21.178s Load: 1.145s, Pack+Encode: 7.573s, Decode+Unpack: 12.459s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4416.9905 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00524-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst (12/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.146s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,092B, BPFP=0.0381 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,392B, BPFP=0.0383 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,520B, BPFP=0.0492 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,984B, BPFP=0.0495 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 93,944B, BPFP=0.0596 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 93,428B, BPFP=0.0593 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 92,472B, BPFP=0.0587 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 91,912B, BPFP=0.0583 +⌛️ [2/4] FRONTEND: Frontend time: 7.800s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.598s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67015388 6.02030628 + layer.9.1 0.00076871 5.97423170 + layer.19.0 3.22151687 25.36585960 + layer.19.1 3.22388957 25.03436281 + layer.29.0 4.24084786 260.34225707 + layer.29.1 4.24602234 260.90829136 + layer.39.0 7.87160790 17794.11764706 + layer.39.1 9.85764150 18144.33539162 + ------------------------------------------------------------------------------------- + TOTAL 4.41655608 4565.26229344 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 647744 +BPFP 0.0514 bits/point +EBPFP 0.0514 equivalent bits/point +MSE 4565.262293 +---------------------- --------------------------------------------------------- +Time: 21.543s Load: 1.146s, Pack+Encode: 7.800s, Decode+Unpack: 12.598s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4565.2623 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00538-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst (13/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.166s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,360B, BPFP=0.0377 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,040B, BPFP=0.0375 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,696B, BPFP=0.0500 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,820B, BPFP=0.0500 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 101,300B, BPFP=0.0643 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 101,224B, BPFP=0.0643 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 101,932B, BPFP=0.0647 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 102,088B, BPFP=0.0648 +⌛️ [2/4] FRONTEND: Frontend time: 7.573s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.798s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00068237 6.07422510 + layer.9.1 0.00070576 6.07256586 + layer.19.0 0.00823322 27.10858029 + layer.19.1 0.08594799 26.96299409 + layer.29.0 0.12200666 303.46254469 + layer.29.1 0.12451052 299.61770799 + layer.39.0 55.99513528 20326.15144621 + layer.39.1 28.81185256 20304.47318817 + ------------------------------------------------------------------------------------- + TOTAL 10.64363429 5162.49040655 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 682460 +BPFP 0.0541 bits/point +EBPFP 0.0541 equivalent bits/point +MSE 5162.490407 +---------------------- --------------------------------------------------------- +Time: 21.537s Load: 1.166s, Pack+Encode: 7.573s, Decode+Unpack: 12.798s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5162.4904 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00927-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst (14/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.165s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,784B, BPFP=0.0386 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,684B, BPFP=0.0385 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 79,728B, BPFP=0.0506 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,568B, BPFP=0.0505 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 99,180B, BPFP=0.0630 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 98,992B, BPFP=0.0628 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,592B, BPFP=0.0600 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 95,144B, BPFP=0.0604 +⌛️ [2/4] FRONTEND: Frontend time: 7.589s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.509s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14556937 6.02961933 + layer.9.1 0.03327741 6.03864991 + layer.19.0 0.11590617 26.72919544 + layer.19.1 0.11733878 26.60384506 + layer.29.0 0.11334742 290.43595629 + layer.29.1 4.29039579 293.88956370 + layer.39.0 9.10722066 18015.19922002 + layer.39.1 44.52401893 18564.34189145 + ------------------------------------------------------------------------------------- + TOTAL 7.30588432 4653.65849265 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 668672 +BPFP 0.0531 bits/point +EBPFP 0.0531 equivalent bits/point +MSE 4653.658493 +---------------------- --------------------------------------------------------- +Time: 21.263s Load: 1.165s, Pack+Encode: 7.589s, Decode+Unpack: 12.509s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4653.6585 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00933-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst (15/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.166s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,248B, BPFP=0.0370 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,536B, BPFP=0.0372 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,880B, BPFP=0.0501 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,240B, BPFP=0.0497 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 99,508B, BPFP=0.0632 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,228B, BPFP=0.0630 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 95,976B, BPFP=0.0609 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 95,716B, BPFP=0.0608 +⌛️ [2/4] FRONTEND: Frontend time: 7.577s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.059s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11318631 6.18402157 + layer.9.1 0.11319129 6.18451223 + layer.19.0 0.00665199 25.46161541 + layer.19.1 0.00853768 25.79009841 + layer.29.0 4.27225940 280.04363016 + layer.29.1 4.27324961 284.42815648 + layer.39.0 14.80262837 18215.87130322 + layer.39.1 16.56649765 18260.62398440 + ------------------------------------------------------------------------------------- + TOTAL 5.01952529 4638.07341524 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 664332 +BPFP 0.0527 bits/point +EBPFP 0.0527 equivalent bits/point +MSE 4638.073415 +---------------------- --------------------------------------------------------- +Time: 20.802s Load: 1.166s, Pack+Encode: 7.577s, Decode+Unpack: 12.059s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4638.0734 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01095-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst (16/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.001s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,284B, BPFP=0.0370 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,460B, BPFP=0.0371 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,536B, BPFP=0.0492 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,984B, BPFP=0.0495 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 98,220B, BPFP=0.0623 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 98,036B, BPFP=0.0622 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,912B, BPFP=0.0602 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 95,304B, BPFP=0.0605 +⌛️ [2/4] FRONTEND: Frontend time: 7.434s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.691s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00065117 6.02511134 + layer.9.1 0.00066201 6.00748116 + layer.19.0 0.00984582 26.02923911 + layer.19.1 0.01156107 25.88728622 + layer.29.0 4.26547583 278.67098635 + layer.29.1 4.26296603 278.91095223 + layer.39.0 11.21169412 17655.67110822 + layer.39.1 9.31977106 17915.91290218 + ------------------------------------------------------------------------------------- + TOTAL 3.63532839 4524.13938335 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 658736 +BPFP 0.0523 bits/point +EBPFP 0.0523 equivalent bits/point +MSE 4524.139383 +---------------------- --------------------------------------------------------- +Time: 21.126s Load: 1.001s, Pack+Encode: 7.434s, Decode+Unpack: 12.691s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4524.1394 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01125-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst (17/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.005s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,244B, BPFP=0.0370 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,000B, BPFP=0.0368 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,108B, BPFP=0.0496 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,760B, BPFP=0.0494 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,776B, BPFP=0.0621 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 97,096B, BPFP=0.0616 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,236B, BPFP=0.0592 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,268B, BPFP=0.0586 +⌛️ [2/4] FRONTEND: Frontend time: 7.595s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.356s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00085355 5.99949157 + layer.9.1 0.00085581 6.00086389 + layer.19.0 0.00808159 25.20698428 + layer.19.1 0.00635426 24.90829136 + layer.29.0 4.24551200 290.26151690 + layer.29.1 4.24803037 287.61567680 + layer.39.0 9.19283951 17548.34839129 + layer.39.1 9.46657027 17135.34481638 + ------------------------------------------------------------------------------------- + TOTAL 3.39613717 4415.46075406 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 652488 +BPFP 0.0518 bits/point +EBPFP 0.0518 equivalent bits/point +MSE 4415.460754 +---------------------- --------------------------------------------------------- +Time: 20.956s Load: 1.005s, Pack+Encode: 7.595s, Decode+Unpack: 12.356s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4415.4608 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01135-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst (18/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.025s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,252B, BPFP=0.0370 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,912B, BPFP=0.0368 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,016B, BPFP=0.0489 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,700B, BPFP=0.0493 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 94,468B, BPFP=0.0600 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,156B, BPFP=0.0604 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 92,140B, BPFP=0.0585 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 93,080B, BPFP=0.0591 +⌛️ [2/4] FRONTEND: Frontend time: 7.550s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.333s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67056234 5.92391636 + layer.9.1 2.67147828 5.95041398 + layer.19.0 0.00618387 26.23169128 + layer.19.1 0.08383032 26.32036328 + layer.29.0 4.28489822 276.88032174 + layer.29.1 4.28470970 277.26541680 + layer.39.0 10.15376305 17352.31459214 + layer.39.1 8.47863686 17877.41696458 + ------------------------------------------------------------------------------------- + TOTAL 4.07925783 4481.03796002 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 645724 +BPFP 0.0512 bits/point +EBPFP 0.0512 equivalent bits/point +MSE 4481.037960 +---------------------- --------------------------------------------------------- +Time: 20.908s Load: 1.025s, Pack+Encode: 7.550s, Decode+Unpack: 12.333s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4481.0380 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01161-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst (19/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 0.909s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,524B, BPFP=0.0371 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,740B, BPFP=0.0373 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,668B, BPFP=0.0499 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,072B, BPFP=0.0502 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 100,944B, BPFP=0.0641 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 101,356B, BPFP=0.0643 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 97,104B, BPFP=0.0616 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 97,624B, BPFP=0.0620 +⌛️ [2/4] FRONTEND: Frontend time: 7.839s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.519s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67132404 5.91101443 + layer.9.1 2.67117709 5.91511365 + layer.19.0 0.00597838 26.18765996 + layer.19.1 0.00605309 26.22587443 + layer.29.0 4.29273040 293.78438820 + layer.29.1 4.29206328 293.88913715 + layer.39.0 9.96127074 17744.05849854 + layer.39.1 10.21295854 17898.96132597 + ------------------------------------------------------------------------------------- + TOTAL 4.26419445 4536.86662654 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 672032 +BPFP 0.0533 bits/point +EBPFP 0.0533 equivalent bits/point +MSE 4536.866627 +---------------------- --------------------------------------------------------- +Time: 21.266s Load: 0.909s, Pack+Encode: 7.839s, Decode+Unpack: 12.519s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4536.8666 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01162-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst (20/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.167s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,884B, BPFP=0.0380 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,116B, BPFP=0.0382 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 79,060B, BPFP=0.0502 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,336B, BPFP=0.0504 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 96,552B, BPFP=0.0613 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 97,172B, BPFP=0.0617 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 92,696B, BPFP=0.0588 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 93,332B, BPFP=0.0592 +⌛️ [2/4] FRONTEND: Frontend time: 7.657s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.359s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14568686 5.98364377 + layer.9.1 0.14558674 5.97521556 + layer.19.0 0.00960369 26.07912283 + layer.19.1 0.03847206 26.00668265 + layer.29.0 4.24438723 285.86031443 + layer.29.1 4.24578970 285.92378941 + layer.39.0 9.23757985 16595.69840754 + layer.39.1 9.43674592 16340.10529737 + ------------------------------------------------------------------------------------- + TOTAL 3.43798151 4196.45405919 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 658148 +BPFP 0.0522 bits/point +EBPFP 0.0522 equivalent bits/point +MSE 4196.454059 +---------------------- --------------------------------------------------------- +Time: 21.183s Load: 1.167s, Pack+Encode: 7.657s, Decode+Unpack: 12.359s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4196.4541 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01163-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst (21/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.142s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,620B, BPFP=0.0372 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,304B, BPFP=0.0376 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,624B, BPFP=0.0493 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,756B, BPFP=0.0500 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 98,028B, BPFP=0.0622 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 98,952B, BPFP=0.0628 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 96,984B, BPFP=0.0616 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 98,408B, BPFP=0.0625 +⌛️ [2/4] FRONTEND: Frontend time: 7.689s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.695s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11307967 6.06193825 + layer.9.1 0.00073224 6.06371301 + layer.19.0 0.08207503 26.96496181 + layer.19.1 0.08214869 26.93820076 + layer.29.0 4.26728487 295.65790543 + layer.29.1 4.26774951 291.27411033 + layer.39.0 12.81553410 19296.21969451 + layer.39.1 23.05196315 19458.02534937 + ------------------------------------------------------------------------------------- + TOTAL 5.58507091 4925.90073418 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 666676 +BPFP 0.0529 bits/point +EBPFP 0.0529 equivalent bits/point +MSE 4925.900734 +---------------------- --------------------------------------------------------- +Time: 21.527s Load: 1.142s, Pack+Encode: 7.689s, Decode+Unpack: 12.695s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4925.9007 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01195-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst (22/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.166s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 61,360B, BPFP=0.0389 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 61,072B, BPFP=0.0388 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,776B, BPFP=0.0500 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,536B, BPFP=0.0499 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 96,044B, BPFP=0.0610 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,264B, BPFP=0.0605 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,728B, BPFP=0.0595 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,972B, BPFP=0.0590 +⌛️ [2/4] FRONTEND: Frontend time: 7.819s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.159s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14519844 6.31808834 + layer.9.1 0.14499054 6.32162453 + layer.19.0 0.12156012 28.20653741 + layer.19.1 0.12030756 28.41208970 + layer.29.0 0.12020218 290.82529656 + layer.29.1 0.12115470 289.45706045 + layer.39.0 8.85439666 19391.18492038 + layer.39.1 8.75438231 19503.94410140 + ------------------------------------------------------------------------------------- + TOTAL 2.29777406 4943.08371484 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 657752 +BPFP 0.0522 bits/point +EBPFP 0.0522 equivalent bits/point +MSE 4943.083715 +---------------------- --------------------------------------------------------- +Time: 21.144s Load: 1.166s, Pack+Encode: 7.819s, Decode+Unpack: 12.159s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4943.0837 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00083-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst (23/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.149s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 61,944B, BPFP=0.0393 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 62,644B, BPFP=0.0398 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 80,820B, BPFP=0.0513 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 81,744B, BPFP=0.0519 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 98,604B, BPFP=0.0626 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,628B, BPFP=0.0632 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,124B, BPFP=0.0591 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,024B, BPFP=0.0597 +⌛️ [2/4] FRONTEND: Frontend time: 7.821s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.769s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14541288 6.35642849 + layer.9.1 0.14479464 6.35338868 + layer.19.0 0.11855170 29.00449403 + layer.19.1 0.11778439 28.94020403 + layer.29.0 0.12648388 299.66137878 + layer.29.1 0.12520221 299.40313211 + layer.39.0 8.37129624 19569.40396490 + layer.39.1 8.45478741 19977.86025349 + ------------------------------------------------------------------------------------- + TOTAL 2.20053917 5027.12290556 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 672532 +BPFP 0.0534 bits/point +EBPFP 0.0534 equivalent bits/point +MSE 5027.122906 +---------------------- --------------------------------------------------------- +Time: 21.739s Load: 1.149s, Pack+Encode: 7.821s, Decode+Unpack: 12.769s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5027.1229 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00084-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst (24/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.167s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,456B, BPFP=0.0365 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,588B, BPFP=0.0366 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 75,024B, BPFP=0.0476 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 74,908B, BPFP=0.0475 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 91,132B, BPFP=0.0578 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 91,252B, BPFP=0.0579 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 89,692B, BPFP=0.0569 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 90,372B, BPFP=0.0574 +⌛️ [2/4] FRONTEND: Frontend time: 7.814s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.219s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14458197 6.27900679 + layer.9.1 0.14461228 6.28391722 + layer.19.0 0.12127609 28.70110903 + layer.19.1 0.12505172 28.40255068 + layer.29.0 0.11568762 293.27567436 + layer.29.1 0.11796058 291.65751950 + layer.39.0 8.63782956 19660.62008450 + layer.39.1 8.69862780 19847.89860253 + ------------------------------------------------------------------------------------- + TOTAL 2.26320345 5020.38980808 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 627424 +BPFP 0.0498 bits/point +EBPFP 0.0498 equivalent bits/point +MSE 5020.389808 +---------------------- --------------------------------------------------------- +Time: 21.201s Load: 1.167s, Pack+Encode: 7.814s, Decode+Unpack: 12.219s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5020.3898 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00085-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst (25/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.200s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 56,740B, BPFP=0.0360 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,180B, BPFP=0.0363 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 75,816B, BPFP=0.0481 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 76,736B, BPFP=0.0487 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 91,324B, BPFP=0.0580 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 91,752B, BPFP=0.0582 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 89,152B, BPFP=0.0566 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 90,104B, BPFP=0.0572 +⌛️ [2/4] FRONTEND: Frontend time: 7.147s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.075s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14478087 6.25945079 + layer.9.1 0.14472154 6.24252011 + layer.19.0 0.13423899 29.49391402 + layer.19.1 0.13534726 28.74936525 + layer.29.0 0.11251127 295.43162983 + layer.29.1 0.11242151 293.84469451 + layer.39.0 10.58490794 19892.99707507 + layer.39.1 8.80008176 19718.93662658 + ------------------------------------------------------------------------------------- + TOTAL 2.52112639 5033.99440952 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 628804 +BPFP 0.0499 bits/point +EBPFP 0.0499 equivalent bits/point +MSE 5033.994410 +---------------------- --------------------------------------------------------- +Time: 20.422s Load: 1.200s, Pack+Encode: 7.147s, Decode+Unpack: 12.075s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5033.9944 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00086-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst (26/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.147s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,376B, BPFP=0.0383 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,280B, BPFP=0.0383 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,028B, BPFP=0.0483 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 76,228B, BPFP=0.0484 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 94,128B, BPFP=0.0597 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 94,248B, BPFP=0.0598 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,584B, BPFP=0.0600 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 95,488B, BPFP=0.0606 +⌛️ [2/4] FRONTEND: Frontend time: 7.795s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.166s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14590075 6.27270563 + layer.9.1 0.14620647 6.27676867 + layer.19.0 0.11628058 27.68961499 + layer.19.1 0.11601873 27.60646531 + layer.29.0 0.11558260 285.83055736 + layer.29.1 0.11828149 286.72617403 + layer.39.0 28.43028163 19308.59538512 + layer.39.1 24.81181701 19421.08287293 + ------------------------------------------------------------------------------------- + TOTAL 6.75004616 4921.26006800 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 651360 +BPFP 0.0517 bits/point +EBPFP 0.0517 equivalent bits/point +MSE 4921.260068 +---------------------- --------------------------------------------------------- +Time: 21.108s Load: 1.147s, Pack+Encode: 7.795s, Decode+Unpack: 12.166s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4921.2601 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00090-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst (27/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.147s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,456B, BPFP=0.0365 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,188B, BPFP=0.0363 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,640B, BPFP=0.0486 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 75,952B, BPFP=0.0482 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 91,668B, BPFP=0.0582 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 91,916B, BPFP=0.0583 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 92,916B, BPFP=0.0590 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,700B, BPFP=0.0588 +⌛️ [2/4] FRONTEND: Frontend time: 7.804s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.160s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11529700 6.14269428 + layer.9.1 0.14629077 6.15166520 + layer.19.0 0.09721754 27.83476702 + layer.19.1 0.12446257 27.93564907 + layer.29.0 4.28687864 278.93861716 + layer.29.1 4.28715508 278.98322229 + layer.39.0 11.34089363 20176.17939552 + layer.39.1 19.75513766 20220.80467988 + ------------------------------------------------------------------------------------- + TOTAL 5.01916661 5127.87133630 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 636436 +BPFP 0.0505 bits/point +EBPFP 0.0505 equivalent bits/point +MSE 5127.871336 +---------------------- --------------------------------------------------------- +Time: 21.111s Load: 1.147s, Pack+Encode: 7.804s, Decode+Unpack: 12.160s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5127.8713 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00116-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst (28/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.144s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,272B, BPFP=0.0376 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,672B, BPFP=0.0379 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,836B, BPFP=0.0494 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,816B, BPFP=0.0494 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 94,424B, BPFP=0.0599 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 94,560B, BPFP=0.0600 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 95,912B, BPFP=0.0609 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 96,588B, BPFP=0.0613 +⌛️ [2/4] FRONTEND: Frontend time: 7.803s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.150s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14538718 6.09172769 + layer.9.1 0.14538559 6.11048899 + layer.19.0 0.11434236 27.63037760 + layer.19.1 0.11406084 27.79340419 + layer.29.0 0.11219077 281.53936464 + layer.29.1 0.11281304 283.85757231 + layer.39.0 79.88316542 20520.49008775 + layer.39.1 46.71980622 20719.73610660 + ------------------------------------------------------------------------------------- + TOTAL 15.91839393 5234.15614122 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 656080 +BPFP 0.0521 bits/point +EBPFP 0.0521 equivalent bits/point +MSE 5234.156141 +---------------------- --------------------------------------------------------- +Time: 21.098s Load: 1.144s, Pack+Encode: 7.803s, Decode+Unpack: 12.150s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5234.1561 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00314-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst (29/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.145s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,616B, BPFP=0.0378 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,152B, BPFP=0.0382 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,060B, BPFP=0.0489 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,660B, BPFP=0.0493 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 92,984B, BPFP=0.0590 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 93,524B, BPFP=0.0594 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 95,120B, BPFP=0.0604 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 95,668B, BPFP=0.0607 +⌛️ [2/4] FRONTEND: Frontend time: 7.690s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.157s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14512126 6.23563942 + layer.9.1 0.14517278 6.22093608 + layer.19.0 0.11689420 28.85952988 + layer.19.1 0.12099910 28.90060327 + layer.29.0 0.11847120 287.77884303 + layer.29.1 0.12399357 285.35649578 + layer.39.0 75.86630139 20562.26584335 + layer.39.1 56.61936342 20490.27624309 + ------------------------------------------------------------------------------------- + TOTAL 16.65703962 5211.98676674 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 651784 +BPFP 0.0517 bits/point +EBPFP 0.0517 equivalent bits/point +MSE 5211.986767 +---------------------- --------------------------------------------------------- +Time: 20.992s Load: 1.145s, Pack+Encode: 7.690s, Decode+Unpack: 12.157s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5211.9868 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00316-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst (30/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.144s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 56,252B, BPFP=0.0357 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 56,796B, BPFP=0.0361 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 74,172B, BPFP=0.0471 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 74,812B, BPFP=0.0475 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 91,048B, BPFP=0.0578 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 92,324B, BPFP=0.0586 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,848B, BPFP=0.0596 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,828B, BPFP=0.0602 +⌛️ [2/4] FRONTEND: Frontend time: 7.712s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.212s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11488649 6.14656689 + layer.9.1 0.14606862 6.12906621 + layer.19.0 0.08767178 27.60568330 + layer.19.1 0.11443626 27.50870115 + layer.29.0 0.10933029 287.72237569 + layer.29.1 0.10817130 283.50641859 + layer.39.0 52.66717785 20663.02762431 + layer.39.1 62.91127214 20687.34611635 + ------------------------------------------------------------------------------------- + TOTAL 14.53237684 5248.62406906 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 634080 +BPFP 0.0503 bits/point +EBPFP 0.0503 equivalent bits/point +MSE 5248.624069 +---------------------- --------------------------------------------------------- +Time: 21.069s Load: 1.144s, Pack+Encode: 7.712s, Decode+Unpack: 12.212s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5248.6241 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00324-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst (31/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.144s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,336B, BPFP=0.0364 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,428B, BPFP=0.0365 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 75,420B, BPFP=0.0479 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 74,760B, BPFP=0.0475 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 91,096B, BPFP=0.0578 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 90,744B, BPFP=0.0576 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,144B, BPFP=0.0591 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 93,064B, BPFP=0.0591 +⌛️ [2/4] FRONTEND: Frontend time: 7.774s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.244s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14509552 6.12399202 + layer.9.1 0.14520687 6.12733017 + layer.19.0 0.12118574 27.36033474 + layer.19.1 0.11709642 27.11819294 + layer.29.0 0.10963326 275.81672489 + layer.29.1 0.10842036 275.04480825 + layer.39.0 53.79489966 19441.62365941 + layer.39.1 62.27410526 19351.24211895 + ------------------------------------------------------------------------------------- + TOTAL 14.60195539 4926.30714517 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 632992 +BPFP 0.0502 bits/point +EBPFP 0.0502 equivalent bits/point +MSE 4926.307145 +---------------------- --------------------------------------------------------- +Time: 21.162s Load: 1.144s, Pack+Encode: 7.774s, Decode+Unpack: 12.244s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4926.3071 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00325-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst (32/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.146s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,144B, BPFP=0.0375 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,128B, BPFP=0.0375 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 75,612B, BPFP=0.0480 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 75,216B, BPFP=0.0477 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 92,316B, BPFP=0.0586 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 91,876B, BPFP=0.0583 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,768B, BPFP=0.0602 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,868B, BPFP=0.0602 +⌛️ [2/4] FRONTEND: Frontend time: 7.528s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.168s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14577075 6.35015082 + layer.9.1 0.14541274 6.36588817 + layer.19.0 0.13069581 28.77792391 + layer.19.1 0.13545482 28.46479424 + layer.29.0 0.11331055 281.28619597 + layer.29.1 0.11244963 280.97528031 + layer.39.0 32.27446072 20123.95060123 + layer.39.1 16.59366367 20191.16022099 + ------------------------------------------------------------------------------------- + TOTAL 6.20640234 5118.41638195 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 642928 +BPFP 0.0510 bits/point +EBPFP 0.0510 equivalent bits/point +MSE 5118.416382 +---------------------- --------------------------------------------------------- +Time: 20.842s Load: 1.146s, Pack+Encode: 7.528s, Decode+Unpack: 12.168s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5118.4164 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00327-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst (33/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.214s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,468B, BPFP=0.0365 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,664B, BPFP=0.0366 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,476B, BPFP=0.0492 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,164B, BPFP=0.0490 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 95,032B, BPFP=0.0603 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,204B, BPFP=0.0604 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,608B, BPFP=0.0601 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,588B, BPFP=0.0600 +⌛️ [2/4] FRONTEND: Frontend time: 7.083s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 11.957s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14585920 6.12149809 + layer.9.1 0.14576220 6.10842414 + layer.19.0 0.12270736 28.00806894 + layer.19.1 0.12453605 27.90010308 + layer.29.0 0.11393550 293.42007231 + layer.29.1 0.11678154 293.74187520 + layer.39.0 53.83016636 20143.17972051 + layer.39.1 40.65720720 19594.56873578 + ------------------------------------------------------------------------------------- + TOTAL 11.90711942 5049.13106226 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 649204 +BPFP 0.0515 bits/point +EBPFP 0.0515 equivalent bits/point +MSE 5049.131062 +---------------------- --------------------------------------------------------- +Time: 20.254s Load: 1.214s, Pack+Encode: 7.083s, Decode+Unpack: 11.957s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5049.1311 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00328-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst (34/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.144s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,548B, BPFP=0.0372 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,924B, BPFP=0.0374 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,400B, BPFP=0.0498 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,808B, BPFP=0.0500 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 96,344B, BPFP=0.0612 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,684B, BPFP=0.0607 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,424B, BPFP=0.0593 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,848B, BPFP=0.0589 +⌛️ [2/4] FRONTEND: Frontend time: 7.786s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.188s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14568599 6.03757401 + layer.9.1 0.03329684 6.05106054 + layer.19.0 0.11848472 27.43548657 + layer.19.1 0.11973745 27.51869211 + layer.29.0 0.10886538 302.60474894 + layer.29.1 0.10946879 302.11445808 + layer.39.0 14.08931437 19741.85895353 + layer.39.1 9.95616799 19804.30289243 + ------------------------------------------------------------------------------------- + TOTAL 3.08512769 5027.24048328 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 652980 +BPFP 0.0518 bits/point +EBPFP 0.0518 equivalent bits/point +MSE 5027.240483 +---------------------- --------------------------------------------------------- +Time: 21.118s Load: 1.144s, Pack+Encode: 7.786s, Decode+Unpack: 12.188s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5027.2405 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00332-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst (35/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.152s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,380B, BPFP=0.0371 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,264B, BPFP=0.0370 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,132B, BPFP=0.0496 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,692B, BPFP=0.0493 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 96,228B, BPFP=0.0611 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,748B, BPFP=0.0608 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,672B, BPFP=0.0601 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 93,876B, BPFP=0.0596 +⌛️ [2/4] FRONTEND: Frontend time: 7.841s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.209s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14493624 6.09072415 + layer.9.1 0.14482686 6.07893811 + layer.19.0 0.11946148 27.67117931 + layer.19.1 0.12828579 27.51007983 + layer.29.0 0.10467725 293.23698001 + layer.29.1 0.10613328 292.09796474 + layer.39.0 22.00188902 19236.78518037 + layer.39.1 19.26198661 19227.37341566 + ------------------------------------------------------------------------------------- + TOTAL 5.25152457 4889.60555777 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 652992 +BPFP 0.0518 bits/point +EBPFP 0.0518 equivalent bits/point +MSE 4889.605558 +---------------------- --------------------------------------------------------- +Time: 21.201s Load: 1.152s, Pack+Encode: 7.841s, Decode+Unpack: 12.209s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4889.6056 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00333-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst (36/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.141s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,652B, BPFP=0.0379 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,752B, BPFP=0.0379 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,972B, BPFP=0.0501 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,756B, BPFP=0.0500 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,112B, BPFP=0.0616 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 96,640B, BPFP=0.0613 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 95,172B, BPFP=0.0604 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 95,292B, BPFP=0.0605 +⌛️ [2/4] FRONTEND: Frontend time: 7.751s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.612s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14475438 6.10427859 + layer.9.1 0.14492096 6.11173056 + layer.19.0 0.11744098 27.45656788 + layer.19.1 0.11578254 27.61019764 + layer.29.0 0.11402616 299.28479444 + layer.29.1 0.11062706 297.17571904 + layer.39.0 28.92800668 19388.56158596 + layer.39.1 10.80449708 19366.86252844 + ------------------------------------------------------------------------------------- + TOTAL 5.06000698 4927.39592532 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 661348 +BPFP 0.0525 bits/point +EBPFP 0.0525 equivalent bits/point +MSE 4927.395925 +---------------------- --------------------------------------------------------- +Time: 21.504s Load: 1.141s, Pack+Encode: 7.751s, Decode+Unpack: 12.612s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4927.3959 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00334-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst (37/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.204s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,048B, BPFP=0.0375 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,668B, BPFP=0.0372 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,908B, BPFP=0.0501 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,576B, BPFP=0.0505 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 99,288B, BPFP=0.0630 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 98,172B, BPFP=0.0623 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,676B, BPFP=0.0601 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,688B, BPFP=0.0588 +⌛️ [2/4] FRONTEND: Frontend time: 7.092s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.067s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14546214 5.97245884 + layer.9.1 0.14553630 5.97988161 + layer.19.0 0.04765745 27.02468669 + layer.19.1 0.04191649 26.99642763 + layer.29.0 0.16505912 312.24207832 + layer.29.1 0.15755973 311.02514625 + layer.39.0 42.51041751 17931.01982450 + layer.39.1 31.38856333 17920.15079623 + ------------------------------------------------------------------------------------- + TOTAL 9.32527151 4567.55141251 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 661024 +BPFP 0.0524 bits/point +EBPFP 0.0524 equivalent bits/point +MSE 4567.551413 +---------------------- --------------------------------------------------------- +Time: 20.363s Load: 1.204s, Pack+Encode: 7.092s, Decode+Unpack: 12.067s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4567.5514 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01347-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst (38/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,096B, BPFP=0.0375 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,908B, BPFP=0.0368 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,552B, BPFP=0.0499 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,640B, BPFP=0.0493 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 98,160B, BPFP=0.0623 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 96,900B, BPFP=0.0615 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 95,972B, BPFP=0.0609 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,440B, BPFP=0.0599 +⌛️ [2/4] FRONTEND: Frontend time: 7.107s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.106s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03343166 6.04110386 + layer.9.1 0.03311388 6.02782743 + layer.19.0 0.03842411 27.08089251 + layer.19.1 0.03806642 27.09451170 + layer.29.0 4.26870163 294.90148684 + layer.29.1 4.26552788 296.86565648 + layer.39.0 33.95300821 18840.45758856 + layer.39.1 48.19954501 18850.62723432 + ------------------------------------------------------------------------------------- + TOTAL 11.35372735 4793.63703771 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 658668 +BPFP 0.0523 bits/point +EBPFP 0.0523 equivalent bits/point +MSE 4793.637038 +---------------------- --------------------------------------------------------- +Time: 20.409s Load: 1.196s, Pack+Encode: 7.107s, Decode+Unpack: 12.106s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4793.6370 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01390-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst (39/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.147s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,444B, BPFP=0.0384 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,636B, BPFP=0.0385 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,396B, BPFP=0.0498 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,752B, BPFP=0.0500 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 96,272B, BPFP=0.0611 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 96,408B, BPFP=0.0612 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 96,016B, BPFP=0.0609 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 96,720B, BPFP=0.0614 +⌛️ [2/4] FRONTEND: Frontend time: 7.719s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.629s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14549638 6.07045405 + layer.9.1 0.14520178 6.04856026 + layer.19.0 0.11487435 27.44070168 + layer.19.1 0.11481158 27.20403650 + layer.29.0 0.10827909 287.87252194 + layer.29.1 0.10618535 287.72174602 + layer.39.0 9.83978281 18879.11862203 + layer.39.1 9.67554703 19291.17972051 + ------------------------------------------------------------------------------------- + TOTAL 2.53127230 4851.58204537 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 663644 +BPFP 0.0527 bits/point +EBPFP 0.0527 equivalent bits/point +MSE 4851.582045 +---------------------- --------------------------------------------------------- +Time: 21.495s Load: 1.147s, Pack+Encode: 7.719s, Decode+Unpack: 12.629s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4851.5820 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01413-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst (40/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.165s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,452B, BPFP=0.0371 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,712B, BPFP=0.0373 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,884B, BPFP=0.0488 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,276B, BPFP=0.0491 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 95,640B, BPFP=0.0607 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 96,916B, BPFP=0.0615 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 97,256B, BPFP=0.0617 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 98,128B, BPFP=0.0623 +⌛️ [2/4] FRONTEND: Frontend time: 7.584s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.296s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00092318 6.04029328 + layer.9.1 0.00095285 6.03194124 + layer.19.0 0.08568402 26.84936627 + layer.19.1 0.08404610 27.05525877 + layer.29.0 0.12100375 295.51507150 + layer.29.1 0.12795564 299.02610091 + layer.39.0 12.85620633 17991.91420214 + layer.39.1 12.98640239 18462.52713682 + ------------------------------------------------------------------------------------- + TOTAL 3.28289678 4639.36992137 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 659264 +BPFP 0.0523 bits/point +EBPFP 0.0523 equivalent bits/point +MSE 4639.369921 +---------------------- --------------------------------------------------------- +Time: 21.044s Load: 1.165s, Pack+Encode: 7.584s, Decode+Unpack: 12.296s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4639.3699 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01421-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst (41/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.175s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 56,476B, BPFP=0.0358 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 56,504B, BPFP=0.0359 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 75,136B, BPFP=0.0477 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 74,120B, BPFP=0.0470 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 92,708B, BPFP=0.0588 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 92,448B, BPFP=0.0587 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 91,192B, BPFP=0.0579 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 91,180B, BPFP=0.0579 +⌛️ [2/4] FRONTEND: Frontend time: 7.832s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.764s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00113376 6.04860215 + layer.9.1 0.00100095 6.04881733 + layer.19.0 0.00983371 26.60864885 + layer.19.1 0.00806405 26.70440973 + layer.29.0 4.28365570 281.27918833 + layer.29.1 4.28597952 286.16574586 + layer.39.0 8.41906814 18073.95775106 + layer.39.1 8.59662605 18136.76048099 + ------------------------------------------------------------------------------------- + TOTAL 3.20067024 4605.44670554 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 629764 +BPFP 0.0500 bits/point +EBPFP 0.0500 equivalent bits/point +MSE 4605.446706 +---------------------- --------------------------------------------------------- +Time: 21.771s Load: 1.175s, Pack+Encode: 7.832s, Decode+Unpack: 12.764s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4605.4467 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01422-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst (42/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.165s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,420B, BPFP=0.0364 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,280B, BPFP=0.0364 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 75,024B, BPFP=0.0476 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 75,804B, BPFP=0.0481 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 92,916B, BPFP=0.0590 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 93,136B, BPFP=0.0591 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 91,176B, BPFP=0.0579 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 90,724B, BPFP=0.0576 +⌛️ [2/4] FRONTEND: Frontend time: 7.878s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.204s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14550294 6.10582421 + layer.9.1 0.14526658 6.09861536 + layer.19.0 0.11599200 26.48037608 + layer.19.1 0.11361485 26.47554944 + layer.29.0 4.26439454 281.08415258 + layer.29.1 4.25587461 279.72964738 + layer.39.0 8.37236706 17878.85342866 + layer.39.1 8.35116642 17683.53071173 + ------------------------------------------------------------------------------------- + TOTAL 3.22052237 4523.54478818 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 633480 +BPFP 0.0503 bits/point +EBPFP 0.0503 equivalent bits/point +MSE 4523.544788 +---------------------- --------------------------------------------------------- +Time: 21.247s Load: 1.165s, Pack+Encode: 7.878s, Decode+Unpack: 12.204s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4523.5448 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01441-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst (43/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.139s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,308B, BPFP=0.0364 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,436B, BPFP=0.0365 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,428B, BPFP=0.0485 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,392B, BPFP=0.0491 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 94,552B, BPFP=0.0600 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 94,996B, BPFP=0.0603 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 92,860B, BPFP=0.0589 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,996B, BPFP=0.0590 +⌛️ [2/4] FRONTEND: Frontend time: 7.867s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.190s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00080404 6.04454166 + layer.9.1 0.00082438 6.05941512 + layer.19.0 0.00843097 26.84458279 + layer.19.1 0.00674472 27.12840480 + layer.29.0 4.27713270 283.18855622 + layer.29.1 4.27133426 282.51042005 + layer.39.0 22.97048921 18237.33116672 + layer.39.1 18.06488920 18384.91257719 + ------------------------------------------------------------------------------------- + TOTAL 6.20008118 4656.75245807 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 643968 +BPFP 0.0511 bits/point +EBPFP 0.0511 equivalent bits/point +MSE 4656.752458 +---------------------- --------------------------------------------------------- +Time: 21.195s Load: 1.139s, Pack+Encode: 7.867s, Decode+Unpack: 12.190s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4656.7525 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00350-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst (44/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.152s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,820B, BPFP=0.0367 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,552B, BPFP=0.0365 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,616B, BPFP=0.0493 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 76,644B, BPFP=0.0486 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 95,392B, BPFP=0.0606 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 94,748B, BPFP=0.0601 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,432B, BPFP=0.0593 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,336B, BPFP=0.0586 +⌛️ [2/4] FRONTEND: Frontend time: 7.649s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.271s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14506060 5.97144388 + layer.9.1 0.14523201 5.98385514 + layer.19.0 0.04621643 26.82405498 + layer.19.1 0.04629335 26.69087941 + layer.29.0 4.27940669 283.50735294 + layer.29.1 4.27759670 283.92393159 + layer.39.0 19.91382637 18529.37666558 + layer.39.1 24.01088215 19232.12479688 + ------------------------------------------------------------------------------------- + TOTAL 6.60806429 4799.30037255 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 645540 +BPFP 0.0512 bits/point +EBPFP 0.0512 equivalent bits/point +MSE 4799.300373 +---------------------- --------------------------------------------------------- +Time: 21.072s Load: 1.152s, Pack+Encode: 7.649s, Decode+Unpack: 12.271s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4799.3004 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00351-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst (45/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.230s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,064B, BPFP=0.0381 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,548B, BPFP=0.0378 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 80,184B, BPFP=0.0509 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,456B, BPFP=0.0504 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 99,328B, BPFP=0.0630 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 98,276B, BPFP=0.0624 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 91,972B, BPFP=0.0584 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 90,812B, BPFP=0.0576 +⌛️ [2/4] FRONTEND: Frontend time: 7.113s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 11.952s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66985745 5.96320356 + layer.9.1 2.66884121 5.97918909 + layer.19.0 3.21935619 25.69900420 + layer.19.1 3.21606501 25.80484746 + layer.29.0 4.24164606 289.17677527 + layer.29.1 4.23648681 289.66810205 + layer.39.0 8.06392628 16539.34871628 + layer.39.1 8.17747540 16789.09587260 + ------------------------------------------------------------------------------------- + TOTAL 4.56170680 4246.34196381 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 659640 +BPFP 0.0523 bits/point +EBPFP 0.0523 equivalent bits/point +MSE 4246.341964 +---------------------- --------------------------------------------------------- +Time: 20.295s Load: 1.230s, Pack+Encode: 7.113s, Decode+Unpack: 11.952s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4246.3420 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00358-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst (46/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.199s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,984B, BPFP=0.0368 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,356B, BPFP=0.0370 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 75,988B, BPFP=0.0482 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 76,968B, BPFP=0.0489 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 93,984B, BPFP=0.0597 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 94,504B, BPFP=0.0600 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 92,076B, BPFP=0.0584 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,508B, BPFP=0.0587 +⌛️ [2/4] FRONTEND: Frontend time: 7.096s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.113s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66795352 5.96920004 + layer.9.1 2.66862889 5.96382815 + layer.19.0 3.22250645 25.41085828 + layer.19.1 3.22577319 25.55518260 + layer.29.0 4.25792136 281.49433295 + layer.29.1 4.25014663 278.84262268 + layer.39.0 8.65209937 18000.36139097 + layer.39.1 8.58450170 17681.93565161 + ------------------------------------------------------------------------------------- + TOTAL 4.69119139 4538.19163341 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 642368 +BPFP 0.0510 bits/point +EBPFP 0.0510 equivalent bits/point +MSE 4538.191633 +---------------------- --------------------------------------------------------- +Time: 20.407s Load: 1.199s, Pack+Encode: 7.096s, Decode+Unpack: 12.113s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4538.1916 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00360-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst (47/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.146s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,032B, BPFP=0.0362 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,172B, BPFP=0.0363 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,540B, BPFP=0.0486 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 76,620B, BPFP=0.0486 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 93,896B, BPFP=0.0596 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 94,552B, BPFP=0.0600 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 90,424B, BPFP=0.0574 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 91,000B, BPFP=0.0578 +⌛️ [2/4] FRONTEND: Frontend time: 7.700s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.142s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00092775 6.09232816 + layer.9.1 0.00093166 6.10731016 + layer.19.0 0.08227225 26.95595344 + layer.19.1 0.08381199 26.68460046 + layer.29.0 0.10725604 288.90229932 + layer.29.1 0.10756977 289.27616185 + layer.39.0 7.96294394 17584.75398115 + layer.39.1 7.95922050 17828.40688983 + ------------------------------------------------------------------------------------- + TOTAL 2.03811674 4507.14744055 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 637236 +BPFP 0.0506 bits/point +EBPFP 0.0506 equivalent bits/point +MSE 4507.147441 +---------------------- --------------------------------------------------------- +Time: 20.987s Load: 1.146s, Pack+Encode: 7.700s, Decode+Unpack: 12.142s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4507.1474 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00361-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst (48/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.174s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,300B, BPFP=0.0370 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,388B, BPFP=0.0371 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,516B, BPFP=0.0498 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,920B, BPFP=0.0495 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 98,920B, BPFP=0.0628 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,236B, BPFP=0.0630 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 95,824B, BPFP=0.0608 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 95,368B, BPFP=0.0605 +⌛️ [2/4] FRONTEND: Frontend time: 7.669s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.295s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66308118 6.02460290 + layer.9.1 2.66351027 6.01806688 + layer.19.0 3.21594155 25.27243713 + layer.19.1 3.21498593 25.16603276 + layer.29.0 4.33566519 304.11106597 + layer.29.1 4.34101296 303.15729607 + layer.39.0 8.65310735 19129.21286968 + layer.39.1 8.66575030 19074.55963601 + ------------------------------------------------------------------------------------- + TOTAL 4.71913184 4859.19025093 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 662472 +BPFP 0.0526 bits/point +EBPFP 0.0526 equivalent bits/point +MSE 4859.190251 +---------------------- --------------------------------------------------------- +Time: 21.138s Load: 1.174s, Pack+Encode: 7.669s, Decode+Unpack: 12.295s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4859.1903 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00384-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst (49/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.163s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,412B, BPFP=0.0371 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,268B, BPFP=0.0370 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 81,220B, BPFP=0.0516 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 81,212B, BPFP=0.0515 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 103,248B, BPFP=0.0655 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 103,864B, BPFP=0.0659 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 101,608B, BPFP=0.0645 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 101,736B, BPFP=0.0646 +⌛️ [2/4] FRONTEND: Frontend time: 7.684s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.478s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.65959259 6.00012124 + layer.9.1 2.65993726 6.01063396 + layer.19.0 3.20866700 25.61171850 + layer.19.1 3.21007805 25.50504499 + layer.29.0 4.27255361 280.12079542 + layer.29.1 4.27602442 283.65099935 + layer.39.0 19.11658068 19112.60058499 + layer.39.1 9.60360322 19084.51218720 + ------------------------------------------------------------------------------------- + TOTAL 6.12587960 4853.00151070 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 689568 +BPFP 0.0547 bits/point +EBPFP 0.0547 equivalent bits/point +MSE 4853.001511 +---------------------- --------------------------------------------------------- +Time: 21.325s Load: 1.163s, Pack+Encode: 7.684s, Decode+Unpack: 12.478s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4853.0015 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00385-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst (50/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,156B, BPFP=0.0382 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,224B, BPFP=0.0382 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,800B, BPFP=0.0500 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,392B, BPFP=0.0498 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,228B, BPFP=0.0617 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 97,856B, BPFP=0.0621 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,180B, BPFP=0.0598 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 95,052B, BPFP=0.0603 +⌛️ [2/4] FRONTEND: Frontend time: 7.097s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 11.962s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67078767 6.06339690 + layer.9.1 2.67131261 6.05562058 + layer.19.0 3.30595795 26.27236858 + layer.19.1 3.30543206 26.11506490 + layer.29.0 0.11228124 301.79854566 + layer.29.1 0.11507649 301.58756500 + layer.39.0 11.41791162 18005.73155671 + layer.39.1 11.38150745 18028.36529087 + ------------------------------------------------------------------------------------- + TOTAL 4.37253339 4587.74867615 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 661888 +BPFP 0.0525 bits/point +EBPFP 0.0525 equivalent bits/point +MSE 4587.748676 +---------------------- --------------------------------------------------------- +Time: 20.255s Load: 1.196s, Pack+Encode: 7.097s, Decode+Unpack: 11.962s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4587.7487 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00557-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst (51/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.144s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,628B, BPFP=0.0378 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,308B, BPFP=0.0376 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 80,048B, BPFP=0.0508 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 80,112B, BPFP=0.0509 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 100,012B, BPFP=0.0635 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,332B, BPFP=0.0631 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 98,744B, BPFP=0.0627 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 98,092B, BPFP=0.0623 +⌛️ [2/4] FRONTEND: Frontend time: 7.771s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.418s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14463971 6.01060349 + layer.9.1 0.14470460 5.99327673 + layer.19.0 0.12255537 27.48845771 + layer.19.1 0.11825690 27.58665096 + layer.29.0 0.11949990 297.53282418 + layer.29.1 0.11467140 293.38966526 + layer.39.0 10.68243977 18779.33571661 + layer.39.1 10.40156301 18881.40396490 + ------------------------------------------------------------------------------------- + TOTAL 2.73104133 4789.84264498 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 675276 +BPFP 0.0536 bits/point +EBPFP 0.0536 equivalent bits/point +MSE 4789.842645 +---------------------- --------------------------------------------------------- +Time: 21.333s Load: 1.144s, Pack+Encode: 7.771s, Decode+Unpack: 12.418s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4789.8426 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00001-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst (52/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.145s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,532B, BPFP=0.0378 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,992B, BPFP=0.0381 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,180B, BPFP=0.0496 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,752B, BPFP=0.0500 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 98,144B, BPFP=0.0623 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,256B, BPFP=0.0630 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 95,836B, BPFP=0.0608 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 97,480B, BPFP=0.0619 +⌛️ [2/4] FRONTEND: Frontend time: 7.764s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.304s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14485273 6.07135095 + layer.9.1 0.14484227 6.08539098 + layer.19.0 0.11969613 27.77840632 + layer.19.1 0.11916645 27.58072494 + layer.29.0 0.11480527 296.10342866 + layer.29.1 0.11451660 296.16605054 + layer.39.0 11.00270276 18121.43776406 + layer.39.1 11.01557422 17869.79785505 + ------------------------------------------------------------------------------------- + TOTAL 2.84701956 4581.37762144 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 667172 +BPFP 0.0529 bits/point +EBPFP 0.0529 equivalent bits/point +MSE 4581.377621 +---------------------- --------------------------------------------------------- +Time: 21.214s Load: 1.145s, Pack+Encode: 7.764s, Decode+Unpack: 12.304s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4581.3776 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00126-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst (53/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.166s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,800B, BPFP=0.0380 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,944B, BPFP=0.0380 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,028B, BPFP=0.0489 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,072B, BPFP=0.0489 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 93,660B, BPFP=0.0595 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 94,080B, BPFP=0.0597 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 92,448B, BPFP=0.0587 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,992B, BPFP=0.0590 +⌛️ [2/4] FRONTEND: Frontend time: 7.590s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.375s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14481488 6.02358349 + layer.9.1 0.14470567 6.02937939 + layer.19.0 0.03819180 26.21872461 + layer.19.1 0.04002141 26.26121984 + layer.29.0 0.11241068 286.95600422 + layer.29.1 0.11133552 288.58461976 + layer.39.0 31.78807483 17721.69905752 + layer.39.1 43.50691623 17945.35326617 + ------------------------------------------------------------------------------------- + TOTAL 9.48580888 4538.39073188 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 647024 +BPFP 0.0513 bits/point +EBPFP 0.0513 equivalent bits/point +MSE 4538.390732 +---------------------- --------------------------------------------------------- +Time: 21.131s Load: 1.166s, Pack+Encode: 7.590s, Decode+Unpack: 12.375s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4538.3907 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00127-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst (54/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.078s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,836B, BPFP=0.0380 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,080B, BPFP=0.0381 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,668B, BPFP=0.0499 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,744B, BPFP=0.0500 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 100,552B, BPFP=0.0638 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,720B, BPFP=0.0633 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 102,988B, BPFP=0.0654 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 102,236B, BPFP=0.0649 +⌛️ [2/4] FRONTEND: Frontend time: 7.540s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.537s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14528272 6.06472353 + layer.9.1 0.14516892 6.03118906 + layer.19.0 0.11319376 26.84931295 + layer.19.1 0.11666145 26.90199210 + layer.29.0 0.21118872 308.90652421 + layer.29.1 0.20646930 305.99268768 + layer.39.0 14.37750853 20709.64445889 + layer.39.1 21.76644002 21064.08839779 + ------------------------------------------------------------------------------------- + TOTAL 4.63523918 5306.80991078 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 682824 +BPFP 0.0542 bits/point +EBPFP 0.0542 equivalent bits/point +MSE 5306.809911 +---------------------- --------------------------------------------------------- +Time: 21.156s Load: 1.078s, Pack+Encode: 7.540s, Decode+Unpack: 12.537s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5306.8099 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00128-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst (55/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.154s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,904B, BPFP=0.0374 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,988B, BPFP=0.0374 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,916B, BPFP=0.0501 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,988B, BPFP=0.0501 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,012B, BPFP=0.0616 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 96,720B, BPFP=0.0614 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,736B, BPFP=0.0601 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,888B, BPFP=0.0602 +⌛️ [2/4] FRONTEND: Frontend time: 7.511s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.633s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14454765 5.98103177 + layer.9.1 0.14475082 5.99928781 + layer.19.0 0.04087094 27.34372969 + layer.19.1 0.11687931 27.47029625 + layer.29.0 0.10817139 292.68386415 + layer.29.1 0.10802081 292.14584010 + layer.39.0 19.80422286 18194.91192720 + layer.39.1 34.29222355 18123.02762431 + ------------------------------------------------------------------------------------- + TOTAL 6.84496092 4621.19545016 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 659152 +BPFP 0.0523 bits/point +EBPFP 0.0523 equivalent bits/point +MSE 4621.195450 +---------------------- --------------------------------------------------------- +Time: 21.297s Load: 1.154s, Pack+Encode: 7.511s, Decode+Unpack: 12.633s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4621.1955 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00131-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst (56/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.122s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,140B, BPFP=0.0375 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,940B, BPFP=0.0374 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 79,452B, BPFP=0.0504 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,412B, BPFP=0.0504 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 96,636B, BPFP=0.0613 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 96,448B, BPFP=0.0612 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,596B, BPFP=0.0600 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,100B, BPFP=0.0597 +⌛️ [2/4] FRONTEND: Frontend time: 7.690s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.340s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14462205 5.93031209 + layer.9.1 0.14495783 5.95077452 + layer.19.0 0.04322015 26.87471563 + layer.19.1 0.03788725 27.04977454 + layer.29.0 0.10021623 290.10858791 + layer.29.1 0.10137775 292.29411765 + layer.39.0 58.66958482 17582.60383490 + layer.39.1 72.48303949 17745.61065973 + ------------------------------------------------------------------------------------- + TOTAL 16.46561320 4497.05284712 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 658724 +BPFP 0.0523 bits/point +EBPFP 0.0523 equivalent bits/point +MSE 4497.052847 +---------------------- --------------------------------------------------------- +Time: 21.151s Load: 1.122s, Pack+Encode: 7.690s, Decode+Unpack: 12.340s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4497.0528 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00132-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst (57/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.164s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,976B, BPFP=0.0387 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 61,380B, BPFP=0.0390 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 79,076B, BPFP=0.0502 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,592B, BPFP=0.0505 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 99,864B, BPFP=0.0634 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,160B, BPFP=0.0629 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 98,624B, BPFP=0.0626 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 98,712B, BPFP=0.0627 +⌛️ [2/4] FRONTEND: Frontend time: 7.654s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.249s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14527927 6.11567235 + layer.9.1 0.14528875 6.10564902 + layer.19.0 0.12591341 27.37005149 + layer.19.1 0.13556211 27.26465764 + layer.29.0 0.11238900 295.44081085 + layer.29.1 0.11028371 293.27551186 + layer.39.0 11.48751193 19159.05882353 + layer.39.1 11.29491489 19075.89210270 + ------------------------------------------------------------------------------------- + TOTAL 2.94464289 4861.31540993 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 677384 +BPFP 0.0537 bits/point +EBPFP 0.0537 equivalent bits/point +MSE 4861.315410 +---------------------- --------------------------------------------------------- +Time: 21.067s Load: 1.164s, Pack+Encode: 7.654s, Decode+Unpack: 12.249s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4861.3154 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00136-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst (58/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.073s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,848B, BPFP=0.0380 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,036B, BPFP=0.0381 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,976B, BPFP=0.0501 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,568B, BPFP=0.0505 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 99,188B, BPFP=0.0630 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 98,636B, BPFP=0.0626 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 97,368B, BPFP=0.0618 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 96,600B, BPFP=0.0613 +⌛️ [2/4] FRONTEND: Frontend time: 7.599s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.338s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14489184 6.08730602 + layer.9.1 0.14511764 6.07911204 + layer.19.0 0.03976490 27.16088621 + layer.19.1 0.11370806 27.02807117 + layer.29.0 0.10933599 292.67515031 + layer.29.1 0.11012027 291.96630240 + layer.39.0 9.10787636 17920.46018850 + layer.39.1 9.00026152 17968.31979201 + ------------------------------------------------------------------------------------- + TOTAL 2.34638457 4567.47210108 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 670220 +BPFP 0.0532 bits/point +EBPFP 0.0532 equivalent bits/point +MSE 4567.472101 +---------------------- --------------------------------------------------------- +Time: 21.010s Load: 1.073s, Pack+Encode: 7.599s, Decode+Unpack: 12.338s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4567.4721 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00193-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst (59/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.180s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 62,744B, BPFP=0.0398 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 62,308B, BPFP=0.0395 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 80,924B, BPFP=0.0514 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 80,112B, BPFP=0.0509 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 98,936B, BPFP=0.0628 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 98,736B, BPFP=0.0627 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,560B, BPFP=0.0600 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,396B, BPFP=0.0599 +⌛️ [2/4] FRONTEND: Frontend time: 7.787s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.167s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00083307 6.07912537 + layer.9.1 0.00247171 6.05723919 + layer.19.0 0.00642632 25.55888700 + layer.19.1 0.00641681 25.46590886 + layer.29.0 0.10256791 281.61122847 + layer.29.1 0.10162673 283.30134872 + layer.39.0 8.50517638 17589.85245369 + layer.39.1 8.55767781 17590.14364641 + ------------------------------------------------------------------------------------- + TOTAL 2.16039959 4476.00872971 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 672716 +BPFP 0.0534 bits/point +EBPFP 0.0534 equivalent bits/point +MSE 4476.008730 +---------------------- --------------------------------------------------------- +Time: 21.134s Load: 1.180s, Pack+Encode: 7.787s, Decode+Unpack: 12.167s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4476.0087 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00196-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst (60/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.132s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,112B, BPFP=0.0375 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,196B, BPFP=0.0376 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,176B, BPFP=0.0496 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,324B, BPFP=0.0497 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 99,100B, BPFP=0.0629 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 98,916B, BPFP=0.0628 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 97,896B, BPFP=0.0621 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 97,704B, BPFP=0.0620 +⌛️ [2/4] FRONTEND: Frontend time: 7.600s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.432s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00064391 6.04328359 + layer.9.1 0.00065402 6.05457261 + layer.19.0 0.08134466 26.49052192 + layer.19.1 0.08141702 26.48953425 + layer.29.0 0.11551180 286.77002762 + layer.29.1 0.11251285 284.73094735 + layer.39.0 10.61319619 18489.14657134 + layer.39.1 10.43102047 18537.47026324 + ------------------------------------------------------------------------------------- + TOTAL 2.67953762 4707.89946524 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 668424 +BPFP 0.0530 bits/point +EBPFP 0.0530 equivalent bits/point +MSE 4707.899465 +---------------------- --------------------------------------------------------- +Time: 21.165s Load: 1.132s, Pack+Encode: 7.600s, Decode+Unpack: 12.432s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4707.8995 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00559-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst (61/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.160s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,652B, BPFP=0.0372 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,556B, BPFP=0.0372 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,204B, BPFP=0.0484 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 76,464B, BPFP=0.0485 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 95,596B, BPFP=0.0607 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,464B, BPFP=0.0606 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 97,284B, BPFP=0.0618 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 97,476B, BPFP=0.0619 +⌛️ [2/4] FRONTEND: Frontend time: 7.622s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.012s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14441776 5.96966785 + layer.9.1 0.14449203 5.97531839 + layer.19.0 0.11315974 26.54743104 + layer.19.1 0.11435745 26.33165167 + layer.29.0 0.12811458 299.19966688 + layer.29.1 0.12952277 300.80287212 + layer.39.0 31.10682331 19969.28956776 + layer.39.1 16.99297713 19595.73610660 + ------------------------------------------------------------------------------------- + TOTAL 6.10923309 5028.73153529 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 655696 +BPFP 0.0520 bits/point +EBPFP 0.0520 equivalent bits/point +MSE 5028.731535 +---------------------- --------------------------------------------------------- +Time: 20.794s Load: 1.160s, Pack+Encode: 7.622s, Decode+Unpack: 12.012s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5028.7315 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00570-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst (62/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.138s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,700B, BPFP=0.0366 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,840B, BPFP=0.0367 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,084B, BPFP=0.0496 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,948B, BPFP=0.0495 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 98,096B, BPFP=0.0623 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 97,468B, BPFP=0.0619 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 98,872B, BPFP=0.0628 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 99,140B, BPFP=0.0629 +⌛️ [2/4] FRONTEND: Frontend time: 7.455s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.723s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00078516 6.01804213 + layer.9.1 0.00079184 6.01770761 + layer.19.0 3.22632161 25.93383876 + layer.19.1 3.22513146 25.83239560 + layer.29.0 0.10494786 282.27421189 + layer.29.1 0.10251782 282.30189714 + layer.39.0 10.88842496 18280.62138447 + layer.39.1 10.78217420 18523.39161521 + ------------------------------------------------------------------------------------- + TOTAL 3.54138686 4679.04888660 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 665148 +BPFP 0.0528 bits/point +EBPFP 0.0528 equivalent bits/point +MSE 4679.048887 +---------------------- --------------------------------------------------------- +Time: 21.317s Load: 1.138s, Pack+Encode: 7.455s, Decode+Unpack: 12.723s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4679.0489 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00758-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst (63/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.217s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 61,340B, BPFP=0.0389 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 61,228B, BPFP=0.0389 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 81,240B, BPFP=0.0516 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 80,880B, BPFP=0.0513 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 103,884B, BPFP=0.0659 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 104,396B, BPFP=0.0663 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 100,372B, BPFP=0.0637 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 100,952B, BPFP=0.0641 +⌛️ [2/4] FRONTEND: Frontend time: 7.110s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 11.990s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14561636 6.03347734 + layer.9.1 0.14552785 6.02425252 + layer.19.0 0.04069186 26.39836387 + layer.19.1 0.03840616 26.38597355 + layer.29.0 0.11346353 294.69968720 + layer.29.1 0.11182956 297.64116835 + layer.39.0 10.19697364 19260.80597985 + layer.39.1 10.11578978 19194.30874228 + ------------------------------------------------------------------------------------- + TOTAL 2.61353734 4889.03720562 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 694292 +BPFP 0.0551 bits/point +EBPFP 0.0551 equivalent bits/point +MSE 4889.037206 +---------------------- --------------------------------------------------------- +Time: 20.317s Load: 1.217s, Pack+Encode: 7.110s, Decode+Unpack: 11.990s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4889.0372 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00764-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst (64/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.131s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,984B, BPFP=0.0368 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,224B, BPFP=0.0370 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,840B, BPFP=0.0500 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,596B, BPFP=0.0499 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,956B, BPFP=0.0622 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 98,596B, BPFP=0.0626 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 97,672B, BPFP=0.0620 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 97,424B, BPFP=0.0618 +⌛️ [2/4] FRONTEND: Frontend time: 7.742s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.588s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14399445 6.01620199 + layer.9.1 0.14558028 6.00324611 + layer.19.0 0.03837104 26.89612193 + layer.19.1 0.04376782 26.71992048 + layer.29.0 0.11695251 295.31871547 + layer.29.1 0.13128335 292.75552486 + layer.39.0 11.28613757 18832.02989925 + layer.39.1 11.84408769 18856.26779331 + ------------------------------------------------------------------------------------- + TOTAL 2.96877184 4792.75092792 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 665292 +BPFP 0.0528 bits/point +EBPFP 0.0528 equivalent bits/point +MSE 4792.750928 +---------------------- --------------------------------------------------------- +Time: 21.461s Load: 1.131s, Pack+Encode: 7.742s, Decode+Unpack: 12.588s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4792.7509 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00845-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst (65/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.158s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,180B, BPFP=0.0363 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,340B, BPFP=0.0364 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,868B, BPFP=0.0494 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,068B, BPFP=0.0489 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,080B, BPFP=0.0616 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 96,676B, BPFP=0.0614 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 100,064B, BPFP=0.0635 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 98,612B, BPFP=0.0626 +⌛️ [2/4] FRONTEND: Frontend time: 7.745s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.129s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14502686 5.92387510 + layer.9.1 0.03259508 5.93325479 + layer.19.0 0.11326540 26.97888568 + layer.19.1 0.11324834 26.82273724 + layer.29.0 0.12250664 300.03715063 + layer.29.1 0.12058897 302.85487082 + layer.39.0 16.17915050 20182.96912577 + layer.39.1 21.66230805 20460.95937602 + ------------------------------------------------------------------------------------- + TOTAL 4.81108623 5164.05990951 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 661888 +BPFP 0.0525 bits/point +EBPFP 0.0525 equivalent bits/point +MSE 5164.059910 +---------------------- --------------------------------------------------------- +Time: 21.033s Load: 1.158s, Pack+Encode: 7.745s, Decode+Unpack: 12.129s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5164.0599 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00870-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst (66/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.147s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 56,200B, BPFP=0.0357 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 56,580B, BPFP=0.0359 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 74,300B, BPFP=0.0472 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 75,672B, BPFP=0.0480 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 90,900B, BPFP=0.0577 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 91,788B, BPFP=0.0583 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 88,792B, BPFP=0.0564 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 90,468B, BPFP=0.0574 +⌛️ [2/4] FRONTEND: Frontend time: 7.808s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.168s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66733019 6.00914294 + layer.9.1 2.66763138 6.02966503 + layer.19.0 3.22293078 24.93327511 + layer.19.1 3.22376992 24.78640417 + layer.29.0 4.27658332 272.93201576 + layer.29.1 4.27160529 270.83153234 + layer.39.0 7.81683598 17039.64120897 + layer.39.1 9.86231960 16869.70035749 + ------------------------------------------------------------------------------------- + TOTAL 4.75112581 4314.35795022 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 624700 +BPFP 0.0496 bits/point +EBPFP 0.0496 equivalent bits/point +MSE 4314.357950 +---------------------- --------------------------------------------------------- +Time: 21.122s Load: 1.147s, Pack+Encode: 7.808s, Decode+Unpack: 12.168s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4314.3580 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00153-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst (67/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.135s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,952B, BPFP=0.0387 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,348B, BPFP=0.0383 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 79,004B, BPFP=0.0501 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,944B, BPFP=0.0495 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,208B, BPFP=0.0617 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,732B, BPFP=0.0608 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 91,536B, BPFP=0.0581 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 89,892B, BPFP=0.0571 +⌛️ [2/4] FRONTEND: Frontend time: 7.689s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.783s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14549763 6.01625912 + layer.9.1 0.14520254 5.99287557 + layer.19.0 0.04746155 26.23238696 + layer.19.1 0.04383140 26.16570269 + layer.29.0 4.26247378 291.46725707 + layer.29.1 4.25497898 291.11163471 + layer.39.0 7.94138086 16976.70328242 + layer.39.1 7.86439079 17448.12999675 + ------------------------------------------------------------------------------------- + TOTAL 3.08815219 4383.97742441 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 652616 +BPFP 0.0518 bits/point +EBPFP 0.0518 equivalent bits/point +MSE 4383.977424 +---------------------- --------------------------------------------------------- +Time: 21.607s Load: 1.135s, Pack+Encode: 7.689s, Decode+Unpack: 12.783s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4383.9774 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00167-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst (68/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.215s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,980B, BPFP=0.0387 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,828B, BPFP=0.0386 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 79,876B, BPFP=0.0507 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,908B, BPFP=0.0507 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 100,868B, BPFP=0.0640 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,856B, BPFP=0.0634 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 98,004B, BPFP=0.0622 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 97,544B, BPFP=0.0619 +⌛️ [2/4] FRONTEND: Frontend time: 7.073s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.144s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11283708 5.98750749 + layer.9.1 0.11300174 5.98208165 + layer.19.0 3.22718329 25.36667716 + layer.19.1 3.22892155 25.39375051 + layer.29.0 4.26448309 278.27230257 + layer.29.1 4.25758082 275.75889665 + layer.39.0 9.82393946 17861.45466363 + layer.39.1 9.78394007 17781.50666233 + ------------------------------------------------------------------------------------- + TOTAL 4.35148589 4532.46531775 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 677864 +BPFP 0.0538 bits/point +EBPFP 0.0538 equivalent bits/point +MSE 4532.465318 +---------------------- --------------------------------------------------------- +Time: 20.432s Load: 1.215s, Pack+Encode: 7.073s, Decode+Unpack: 12.144s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4532.4653 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00208-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst (69/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.140s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,400B, BPFP=0.0383 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,304B, BPFP=0.0383 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,876B, BPFP=0.0501 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,500B, BPFP=0.0498 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 94,856B, BPFP=0.0602 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 94,876B, BPFP=0.0602 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,464B, BPFP=0.0600 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,088B, BPFP=0.0597 +⌛️ [2/4] FRONTEND: Frontend time: 7.567s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.111s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14503861 6.18459221 + layer.9.1 0.14483112 6.18844578 + layer.19.0 0.11529889 27.39229820 + layer.19.1 0.11517203 27.31550871 + layer.29.0 0.11961639 287.73110985 + layer.29.1 0.11795276 289.83652909 + layer.39.0 83.84633978 18425.53526162 + layer.39.1 174.87768118 18738.74683133 + ------------------------------------------------------------------------------------- + TOTAL 32.43524135 4726.11632210 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 656364 +BPFP 0.0521 bits/point +EBPFP 0.0521 equivalent bits/point +MSE 4726.116322 +---------------------- --------------------------------------------------------- +Time: 20.819s Load: 1.140s, Pack+Encode: 7.567s, Decode+Unpack: 12.111s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4726.1163 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01201-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst (70/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.147s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,820B, BPFP=0.0380 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,700B, BPFP=0.0379 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,816B, BPFP=0.0500 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,184B, BPFP=0.0496 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 98,992B, BPFP=0.0628 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 98,144B, BPFP=0.0623 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 92,160B, BPFP=0.0585 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 91,724B, BPFP=0.0582 +⌛️ [2/4] FRONTEND: Frontend time: 7.805s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.152s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.81453913 5.95349696 + layer.9.1 0.14528001 5.94629065 + layer.19.0 3.26598681 26.56858852 + layer.19.1 0.04116655 26.56151741 + layer.29.0 4.28557138 310.34317111 + layer.29.1 4.28198282 305.07661683 + layer.39.0 74.89367180 18772.76438089 + layer.39.1 42.04871577 18841.96555086 + ------------------------------------------------------------------------------------- + TOTAL 16.47211428 4786.89745165 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 657540 +BPFP 0.0522 bits/point +EBPFP 0.0522 equivalent bits/point +MSE 4786.897452 +---------------------- --------------------------------------------------------- +Time: 21.104s Load: 1.147s, Pack+Encode: 7.805s, Decode+Unpack: 12.152s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4786.8975 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01296-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst (71/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.136s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,488B, BPFP=0.0378 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,156B, BPFP=0.0375 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 79,500B, BPFP=0.0505 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,452B, BPFP=0.0504 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 100,596B, BPFP=0.0639 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,360B, BPFP=0.0631 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 95,936B, BPFP=0.0609 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,680B, BPFP=0.0601 +⌛️ [2/4] FRONTEND: Frontend time: 7.531s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.498s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66710422 5.95097828 + layer.9.1 2.66812426 5.94680543 + layer.19.0 3.22059776 26.65135989 + layer.19.1 3.22546153 26.55558377 + layer.29.0 0.11226317 293.00958726 + layer.29.1 0.11257672 293.02252600 + layer.39.0 59.39237691 18708.73708157 + layer.39.1 37.52358222 18388.07149821 + ------------------------------------------------------------------------------------- + TOTAL 13.61526085 4718.49317755 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 668168 +BPFP 0.0530 bits/point +EBPFP 0.0530 equivalent bits/point +MSE 4718.493178 +---------------------- --------------------------------------------------------- +Time: 21.165s Load: 1.136s, Pack+Encode: 7.531s, Decode+Unpack: 12.498s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4718.4932 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01313-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst (72/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.211s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,804B, BPFP=0.0386 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 60,920B, BPFP=0.0387 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,888B, BPFP=0.0488 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 76,996B, BPFP=0.0489 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 92,960B, BPFP=0.0590 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 93,228B, BPFP=0.0592 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 92,764B, BPFP=0.0589 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,988B, BPFP=0.0590 +⌛️ [2/4] FRONTEND: Frontend time: 7.328s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.377s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14519123 5.93696237 + layer.9.1 0.14511500 5.94236091 + layer.19.0 0.03974548 24.61610843 + layer.19.1 0.03981401 24.64367180 + layer.29.0 4.26343511 265.48419727 + layer.29.1 4.25610090 263.85489113 + layer.39.0 7.90972018 17776.16379591 + layer.39.1 8.05601540 17927.64250894 + ------------------------------------------------------------------------------------- + TOTAL 3.10689216 4536.78556209 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 647548 +BPFP 0.0514 bits/point +EBPFP 0.0514 equivalent bits/point +MSE 4536.785562 +---------------------- --------------------------------------------------------- +Time: 20.916s Load: 1.211s, Pack+Encode: 7.328s, Decode+Unpack: 12.377s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4536.7856 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00032-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst (73/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.202s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,672B, BPFP=0.0379 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,564B, BPFP=0.0378 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,472B, BPFP=0.0485 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,252B, BPFP=0.0490 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 94,876B, BPFP=0.0602 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,844B, BPFP=0.0608 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,760B, BPFP=0.0601 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 95,272B, BPFP=0.0605 +⌛️ [2/4] FRONTEND: Frontend time: 7.145s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.096s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14586543 5.99812304 + layer.9.1 0.14572574 5.98740276 + layer.19.0 0.03953905 25.52491012 + layer.19.1 0.03760033 25.60452043 + layer.29.0 0.10448607 286.89732288 + layer.29.1 0.10697372 283.88379509 + layer.39.0 14.19073468 19449.33116672 + layer.39.1 8.92149669 19403.05752356 + ------------------------------------------------------------------------------------- + TOTAL 2.96155271 4935.78559558 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 653712 +BPFP 0.0519 bits/point +EBPFP 0.0519 equivalent bits/point +MSE 4935.785596 +---------------------- --------------------------------------------------------- +Time: 20.442s Load: 1.202s, Pack+Encode: 7.145s, Decode+Unpack: 12.096s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4935.7856 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00041-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst (74/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.195s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,332B, BPFP=0.0370 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,700B, BPFP=0.0373 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,828B, BPFP=0.0500 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,720B, BPFP=0.0500 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,332B, BPFP=0.0618 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 97,868B, BPFP=0.0621 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 96,088B, BPFP=0.0610 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 95,940B, BPFP=0.0609 +⌛️ [2/4] FRONTEND: Frontend time: 7.107s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.083s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00251883 6.01342559 + layer.9.1 0.14409062 6.03431267 + layer.19.0 0.12740102 26.72273623 + layer.19.1 0.12254588 27.32231577 + layer.29.0 4.25147928 283.85747075 + layer.29.1 4.25065697 286.07056386 + layer.39.0 9.21805114 18428.11959701 + layer.39.1 9.03214690 17926.55443614 + ------------------------------------------------------------------------------------- + TOTAL 3.39361133 4623.83685725 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 661808 +BPFP 0.0525 bits/point +EBPFP 0.0525 equivalent bits/point +MSE 4623.836857 +---------------------- --------------------------------------------------------- +Time: 20.385s Load: 1.195s, Pack+Encode: 7.107s, Decode+Unpack: 12.083s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4623.8369 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00042-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst (75/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.139s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,616B, BPFP=0.0372 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,072B, BPFP=0.0375 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,516B, BPFP=0.0486 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,184B, BPFP=0.0490 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 94,364B, BPFP=0.0599 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 94,688B, BPFP=0.0601 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 92,228B, BPFP=0.0585 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,572B, BPFP=0.0588 +⌛️ [2/4] FRONTEND: Frontend time: 7.844s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.262s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14550905 6.09782827 + layer.9.1 0.14590163 6.09751089 + layer.19.0 0.12839093 27.58482288 + layer.19.1 0.12422524 27.56910901 + layer.29.0 0.11695262 292.33849935 + layer.29.1 0.11389293 290.70535424 + layer.39.0 10.18180439 18803.18882028 + layer.39.1 10.42432323 18852.74098148 + ------------------------------------------------------------------------------------- + TOTAL 2.67262500 4788.29036580 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 645240 +BPFP 0.0512 bits/point +EBPFP 0.0512 equivalent bits/point +MSE 4788.290366 +---------------------- --------------------------------------------------------- +Time: 21.244s Load: 1.139s, Pack+Encode: 7.844s, Decode+Unpack: 12.262s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4788.2904 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00620-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst (76/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,344B, BPFP=0.0370 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,968B, BPFP=0.0368 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,644B, BPFP=0.0493 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,680B, BPFP=0.0493 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,928B, BPFP=0.0622 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 96,772B, BPFP=0.0614 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 97,160B, BPFP=0.0617 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 96,360B, BPFP=0.0612 +⌛️ [2/4] FRONTEND: Frontend time: 7.109s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.072s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14509775 6.01017567 + layer.9.1 0.14508723 5.98333338 + layer.19.0 0.11633494 26.71516239 + layer.19.1 0.11804005 26.90758297 + layer.29.0 0.15409572 296.21051349 + layer.29.1 0.14997486 296.46894296 + layer.39.0 9.23291952 19761.27396815 + layer.39.1 9.22304726 19762.16054599 + ------------------------------------------------------------------------------------- + TOTAL 2.41057467 5022.71627813 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 659856 +BPFP 0.0524 bits/point +EBPFP 0.0524 equivalent bits/point +MSE 5022.716278 +---------------------- --------------------------------------------------------- +Time: 20.377s Load: 1.196s, Pack+Encode: 7.109s, Decode+Unpack: 12.072s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5022.7163 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00636-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst (77/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.190s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,480B, BPFP=0.0371 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,812B, BPFP=0.0373 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,328B, BPFP=0.0491 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,612B, BPFP=0.0493 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 96,508B, BPFP=0.0613 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 96,684B, BPFP=0.0614 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 96,488B, BPFP=0.0612 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 97,264B, BPFP=0.0617 +⌛️ [2/4] FRONTEND: Frontend time: 7.126s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.062s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14497060 6.08004258 + layer.9.1 0.14492971 6.08243114 + layer.19.0 0.11929473 27.60328140 + layer.19.1 0.11869117 27.60153711 + layer.29.0 0.13715227 299.58250731 + layer.29.1 0.14278979 301.60759262 + layer.39.0 9.99110525 20606.58563536 + layer.39.1 10.01170034 20858.64413390 + ------------------------------------------------------------------------------------- + TOTAL 2.60132923 5266.72339518 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 659176 +BPFP 0.0523 bits/point +EBPFP 0.0523 equivalent bits/point +MSE 5266.723395 +---------------------- --------------------------------------------------------- +Time: 20.379s Load: 1.190s, Pack+Encode: 7.126s, Decode+Unpack: 12.062s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5266.7234 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00637-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst (78/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.149s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,384B, BPFP=0.0364 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,124B, BPFP=0.0363 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,080B, BPFP=0.0489 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,624B, BPFP=0.0493 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 96,072B, BPFP=0.0610 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,732B, BPFP=0.0608 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,388B, BPFP=0.0593 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 93,392B, BPFP=0.0593 +⌛️ [2/4] FRONTEND: Frontend time: 7.640s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.589s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14589856 5.99745402 + layer.9.1 0.03321603 5.99128996 + layer.19.0 0.11866178 28.03519053 + layer.19.1 0.11267978 27.63877153 + layer.29.0 0.10803594 298.57143728 + layer.29.1 0.10714094 299.36157377 + layer.39.0 11.58943751 18864.68378291 + layer.39.1 9.70079103 19337.24406890 + ------------------------------------------------------------------------------------- + TOTAL 2.73948269 4858.44044611 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 647796 +BPFP 0.0514 bits/point +EBPFP 0.0514 equivalent bits/point +MSE 4858.440446 +---------------------- --------------------------------------------------------- +Time: 21.378s Load: 1.149s, Pack+Encode: 7.640s, Decode+Unpack: 12.589s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4858.4404 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00410-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst (79/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.161s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,292B, BPFP=0.0376 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,032B, BPFP=0.0375 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,420B, BPFP=0.0498 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,844B, BPFP=0.0500 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 99,556B, BPFP=0.0632 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,448B, BPFP=0.0631 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 97,220B, BPFP=0.0617 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 97,376B, BPFP=0.0618 +⌛️ [2/4] FRONTEND: Frontend time: 7.872s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.412s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03308301 5.94857004 + layer.9.1 0.14566304 5.95974100 + layer.19.0 0.03810260 26.48544900 + layer.19.1 0.03780774 26.65232979 + layer.29.0 0.11592613 293.40146653 + layer.29.1 0.11717217 297.09026649 + layer.39.0 9.98032847 18821.97985050 + layer.39.1 9.70849498 18617.79785505 + ------------------------------------------------------------------------------------- + TOTAL 2.52207227 4761.91444105 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 669188 +BPFP 0.0531 bits/point +EBPFP 0.0531 equivalent bits/point +MSE 4761.914441 +---------------------- --------------------------------------------------------- +Time: 21.444s Load: 1.161s, Pack+Encode: 7.872s, Decode+Unpack: 12.412s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4761.9144 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00411-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst (80/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.143s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 61,340B, BPFP=0.0389 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 61,236B, BPFP=0.0389 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 81,608B, BPFP=0.0518 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 81,608B, BPFP=0.0518 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 103,048B, BPFP=0.0654 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 103,140B, BPFP=0.0655 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 99,472B, BPFP=0.0631 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 99,068B, BPFP=0.0629 +⌛️ [2/4] FRONTEND: Frontend time: 7.770s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.451s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14555504 5.97797990 + layer.9.1 0.14557384 5.96555975 + layer.19.0 0.03995539 27.01036166 + layer.19.1 0.04542811 26.99978165 + layer.29.0 0.12033866 311.46376341 + layer.29.1 0.13252172 312.10013812 + layer.39.0 10.37566776 18655.06402340 + layer.39.1 9.84188447 18462.90932727 + ------------------------------------------------------------------------------------- + TOTAL 2.60586562 4725.93636689 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 690520 +BPFP 0.0548 bits/point +EBPFP 0.0548 equivalent bits/point +MSE 4725.936367 +---------------------- --------------------------------------------------------- +Time: 21.364s Load: 1.143s, Pack+Encode: 7.770s, Decode+Unpack: 12.451s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4725.9364 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00412-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst (81/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,152B, BPFP=0.0369 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,180B, BPFP=0.0369 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 75,376B, BPFP=0.0478 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 75,916B, BPFP=0.0482 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 92,880B, BPFP=0.0590 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 93,320B, BPFP=0.0592 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 91,496B, BPFP=0.0581 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,180B, BPFP=0.0585 +⌛️ [2/4] FRONTEND: Frontend time: 7.059s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 11.922s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14470203 6.00991606 + layer.9.1 0.14481130 5.99466049 + layer.19.0 0.11257574 27.34509821 + layer.19.1 0.11422884 27.47564084 + layer.29.0 0.10456927 286.75430614 + layer.29.1 0.10551051 290.05969695 + layer.39.0 10.36536069 19065.67045824 + layer.39.1 11.81531702 19171.59961001 + ------------------------------------------------------------------------------------- + TOTAL 2.86338443 4860.11367337 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 637500 +BPFP 0.0506 bits/point +EBPFP 0.0506 equivalent bits/point +MSE 4860.113673 +---------------------- --------------------------------------------------------- +Time: 20.177s Load: 1.196s, Pack+Encode: 7.059s, Decode+Unpack: 11.922s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4860.1137 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00413-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst (82/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.132s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,156B, BPFP=0.0369 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,444B, BPFP=0.0371 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 79,092B, BPFP=0.0502 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,140B, BPFP=0.0502 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 96,848B, BPFP=0.0615 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 97,744B, BPFP=0.0620 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,280B, BPFP=0.0598 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 95,104B, BPFP=0.0604 +⌛️ [2/4] FRONTEND: Frontend time: 7.864s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.128s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14530003 6.01573862 + layer.9.1 0.14546206 6.01108781 + layer.19.0 0.11891763 27.59844969 + layer.19.1 0.11677460 27.24506926 + layer.29.0 4.29725807 297.94276081 + layer.29.1 4.29692800 298.00662171 + layer.39.0 11.61914761 19425.21676958 + layer.39.1 11.22064282 19195.64510887 + ------------------------------------------------------------------------------------- + TOTAL 3.99505385 4910.46020079 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 658808 +BPFP 0.0523 bits/point +EBPFP 0.0523 equivalent bits/point +MSE 4910.460201 +---------------------- --------------------------------------------------------- +Time: 21.124s Load: 1.132s, Pack+Encode: 7.864s, Decode+Unpack: 12.128s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4910.4602 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00433-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst (83/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.195s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,584B, BPFP=0.0372 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,464B, BPFP=0.0371 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,384B, BPFP=0.0491 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 77,076B, BPFP=0.0489 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 95,884B, BPFP=0.0609 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,672B, BPFP=0.0607 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 92,576B, BPFP=0.0588 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,372B, BPFP=0.0586 +⌛️ [2/4] FRONTEND: Frontend time: 7.047s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 11.999s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67106792 5.97957312 + layer.9.1 2.67195307 5.97994000 + layer.19.0 0.08237472 26.46528934 + layer.19.1 0.08192194 26.30055401 + layer.29.0 0.11152953 287.55730013 + layer.29.1 0.11703055 286.89925252 + layer.39.0 163.01811830 18625.67175821 + layer.39.1 58.15221299 18653.86675333 + ------------------------------------------------------------------------------------- + TOTAL 28.36327613 4739.84005258 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 648012 +BPFP 0.0514 bits/point +EBPFP 0.0514 equivalent bits/point +MSE 4739.840053 +---------------------- --------------------------------------------------------- +Time: 20.241s Load: 1.195s, Pack+Encode: 7.047s, Decode+Unpack: 11.999s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4739.8401 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00443-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst (84/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,328B, BPFP=0.0377 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,400B, BPFP=0.0377 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,456B, BPFP=0.0498 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,544B, BPFP=0.0499 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,492B, BPFP=0.0619 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 97,728B, BPFP=0.0620 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 96,216B, BPFP=0.0611 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 96,696B, BPFP=0.0614 +⌛️ [2/4] FRONTEND: Frontend time: 7.078s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.055s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14620710 6.07535622 + layer.9.1 0.14642976 6.09581357 + layer.19.0 0.11726453 26.78738676 + layer.19.1 0.11958517 26.63138304 + layer.29.0 0.10693079 291.26015600 + layer.29.1 0.10826971 290.76527462 + layer.39.0 43.01306569 19775.36301592 + layer.39.1 17.12450997 20029.33506662 + ------------------------------------------------------------------------------------- + TOTAL 7.61028284 5056.53918159 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 663860 +BPFP 0.0527 bits/point +EBPFP 0.0527 equivalent bits/point +MSE 5056.539182 +---------------------- --------------------------------------------------------- +Time: 20.326s Load: 1.193s, Pack+Encode: 7.078s, Decode+Unpack: 12.055s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 5056.5392 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00444-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst (85/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.137s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,904B, BPFP=0.0380 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,576B, BPFP=0.0378 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 80,700B, BPFP=0.0512 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,588B, BPFP=0.0505 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 98,724B, BPFP=0.0627 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 97,916B, BPFP=0.0622 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,708B, BPFP=0.0601 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,816B, BPFP=0.0602 +⌛️ [2/4] FRONTEND: Frontend time: 7.643s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.249s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03173930 5.93754253 + layer.9.1 0.03345565 5.93911671 + layer.19.0 3.26068347 26.27497359 + layer.19.1 3.26087326 25.89829786 + layer.29.0 4.24610771 281.21839454 + layer.29.1 4.24089229 277.92636903 + layer.39.0 8.81319124 18180.63178421 + layer.39.1 8.71779153 18510.07734807 + ------------------------------------------------------------------------------------- + TOTAL 4.07559181 4664.23797832 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 665932 +BPFP 0.0528 bits/point +EBPFP 0.0528 equivalent bits/point +MSE 4664.237978 +---------------------- --------------------------------------------------------- +Time: 21.030s Load: 1.137s, Pack+Encode: 7.643s, Decode+Unpack: 12.249s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4664.2380 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00445-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst (86/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.169s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 56,760B, BPFP=0.0360 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 56,824B, BPFP=0.0361 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 75,208B, BPFP=0.0477 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 75,248B, BPFP=0.0478 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 93,576B, BPFP=0.0594 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 93,152B, BPFP=0.0591 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 90,820B, BPFP=0.0576 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 91,080B, BPFP=0.0578 +⌛️ [2/4] FRONTEND: Frontend time: 7.828s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.146s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00078599 5.95570843 + layer.9.1 0.00079117 5.96024054 + layer.19.0 0.00795310 26.06195158 + layer.19.1 0.00811505 25.95449860 + layer.29.0 4.25797468 281.81694833 + layer.29.1 4.25504309 281.93642346 + layer.39.0 81.06806549 18018.82352941 + layer.39.1 44.82015254 18029.72765681 + ------------------------------------------------------------------------------------- + TOTAL 16.80236014 4584.52961965 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 632668 +BPFP 0.0502 bits/point +EBPFP 0.0502 equivalent bits/point +MSE 4584.529620 +---------------------- --------------------------------------------------------- +Time: 21.143s Load: 1.169s, Pack+Encode: 7.828s, Decode+Unpack: 12.146s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4584.5296 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00446-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst (87/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.148s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,928B, BPFP=0.0368 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,132B, BPFP=0.0369 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,160B, BPFP=0.0496 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,100B, BPFP=0.0496 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 96,148B, BPFP=0.0610 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 96,432B, BPFP=0.0612 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,124B, BPFP=0.0591 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 93,444B, BPFP=0.0593 +⌛️ [2/4] FRONTEND: Frontend time: 7.811s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.123s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00082351 5.97605660 + layer.9.1 0.02968625 5.97205133 + layer.19.0 0.00841222 26.79836285 + layer.19.1 0.03743129 26.71309819 + layer.29.0 4.28408194 283.54111147 + layer.29.1 4.28564945 285.05167371 + layer.39.0 8.35370986 18086.67663308 + layer.39.1 8.52557915 18288.35359116 + ------------------------------------------------------------------------------------- + TOTAL 3.19067171 4626.13532230 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 651468 +BPFP 0.0517 bits/point +EBPFP 0.0517 equivalent bits/point +MSE 4626.135322 +---------------------- --------------------------------------------------------- +Time: 21.083s Load: 1.148s, Pack+Encode: 7.811s, Decode+Unpack: 12.123s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4626.1353 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00447-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst (88/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.142s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,668B, BPFP=0.0372 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,636B, BPFP=0.0372 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 79,288B, BPFP=0.0503 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,520B, BPFP=0.0505 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 100,384B, BPFP=0.0637 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 100,224B, BPFP=0.0636 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 97,332B, BPFP=0.0618 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 96,464B, BPFP=0.0612 +⌛️ [2/4] FRONTEND: Frontend time: 7.797s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.115s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14529990 5.97255152 + layer.9.1 0.14524076 5.97211862 + layer.19.0 0.03780325 26.83941339 + layer.19.1 0.03783790 26.98716790 + layer.29.0 4.32098184 290.47519906 + layer.29.1 4.32100596 287.83896653 + layer.39.0 9.32673680 18386.07604810 + layer.39.1 9.31823369 18397.73415665 + ------------------------------------------------------------------------------------- + TOTAL 3.45664251 4678.48695272 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 670516 +BPFP 0.0532 bits/point +EBPFP 0.0532 equivalent bits/point +MSE 4678.486953 +---------------------- --------------------------------------------------------- +Time: 21.055s Load: 1.142s, Pack+Encode: 7.797s, Decode+Unpack: 12.115s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4678.4870 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00461-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst (89/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.145s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,028B, BPFP=0.0381 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,744B, BPFP=0.0379 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 80,260B, BPFP=0.0509 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,840B, BPFP=0.0507 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 100,260B, BPFP=0.0636 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,136B, BPFP=0.0629 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 96,352B, BPFP=0.0612 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,704B, BPFP=0.0601 +⌛️ [2/4] FRONTEND: Frontend time: 7.726s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.315s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14497286 6.02334546 + layer.9.1 0.14497296 5.99394385 + layer.19.0 0.03962668 27.39254194 + layer.19.1 0.11751332 27.58888782 + layer.29.0 0.14529291 299.96591648 + layer.29.1 0.16241527 300.34835067 + layer.39.0 11.40179406 19189.27656809 + layer.39.1 13.03458244 18929.09067273 + ------------------------------------------------------------------------------------- + TOTAL 3.14889631 4848.21002838 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 670324 +BPFP 0.0532 bits/point +EBPFP 0.0532 equivalent bits/point +MSE 4848.210028 +---------------------- --------------------------------------------------------- +Time: 21.187s Load: 1.145s, Pack+Encode: 7.726s, Decode+Unpack: 12.315s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4848.2100 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00462-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst (90/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.201s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,076B, BPFP=0.0362 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,360B, BPFP=0.0364 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,436B, BPFP=0.0498 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,288B, BPFP=0.0497 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 98,208B, BPFP=0.0623 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 98,324B, BPFP=0.0624 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 96,716B, BPFP=0.0614 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 97,172B, BPFP=0.0617 +⌛️ [2/4] FRONTEND: Frontend time: 7.090s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.076s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14513344 6.00968184 + layer.9.1 0.03283094 6.02973866 + layer.19.0 0.11544709 27.84888386 + layer.19.1 0.11326018 27.98226001 + layer.29.0 0.14483232 300.36407215 + layer.29.1 0.14672551 299.50255931 + layer.39.0 10.02784076 18533.93175171 + layer.39.1 15.62606130 18915.96100097 + ------------------------------------------------------------------------------------- + TOTAL 3.29401644 4764.70374356 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 661580 +BPFP 0.0525 bits/point +EBPFP 0.0525 equivalent bits/point +MSE 4764.703744 +---------------------- --------------------------------------------------------- +Time: 20.368s Load: 1.201s, Pack+Encode: 7.090s, Decode+Unpack: 12.076s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4764.7037 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00463-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst (91/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.149s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,656B, BPFP=0.0366 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,852B, BPFP=0.0367 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,820B, BPFP=0.0494 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,076B, BPFP=0.0496 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,444B, BPFP=0.0619 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 97,388B, BPFP=0.0618 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 96,436B, BPFP=0.0612 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 96,648B, BPFP=0.0613 +⌛️ [2/4] FRONTEND: Frontend time: 7.781s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.124s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14492303 6.05822179 + layer.9.1 0.14484742 6.05168196 + layer.19.0 0.11740684 27.74963438 + layer.19.1 0.11489933 27.75619516 + layer.29.0 0.12072669 292.45787293 + layer.29.1 0.12118037 293.03968963 + layer.39.0 10.74778980 19183.07052324 + layer.39.1 11.83662176 19492.23269418 + ------------------------------------------------------------------------------------- + TOTAL 2.91854940 4916.05206416 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 659320 +BPFP 0.0523 bits/point +EBPFP 0.0523 equivalent bits/point +MSE 4916.052064 +---------------------- --------------------------------------------------------- +Time: 21.054s Load: 1.149s, Pack+Encode: 7.781s, Decode+Unpack: 12.124s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4916.0521 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00464-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst (92/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.138s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,840B, BPFP=0.0367 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,064B, BPFP=0.0369 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,400B, BPFP=0.0498 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 79,248B, BPFP=0.0503 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,156B, BPFP=0.0617 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,176B, BPFP=0.0630 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 95,704B, BPFP=0.0607 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 97,124B, BPFP=0.0616 +⌛️ [2/4] FRONTEND: Frontend time: 7.816s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.109s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14465000 6.09864202 + layer.9.1 0.14489275 6.08986787 + layer.19.0 0.11978787 28.14525613 + layer.19.1 0.12819003 28.31053735 + layer.29.0 0.12519148 294.93376259 + layer.29.1 0.13018718 295.82566217 + layer.39.0 10.77894586 19230.31784205 + layer.39.1 10.25834823 19104.44198895 + ------------------------------------------------------------------------------------- + TOTAL 2.72877418 4874.27044489 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 662712 +BPFP 0.0526 bits/point +EBPFP 0.0526 equivalent bits/point +MSE 4874.270445 +---------------------- --------------------------------------------------------- +Time: 21.063s Load: 1.138s, Pack+Encode: 7.816s, Decode+Unpack: 12.109s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4874.2704 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00465-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst (93/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.140s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 60,660B, BPFP=0.0385 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,948B, BPFP=0.0374 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 79,504B, BPFP=0.0505 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,308B, BPFP=0.0497 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 97,696B, BPFP=0.0620 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,776B, BPFP=0.0608 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,960B, BPFP=0.0596 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 91,792B, BPFP=0.0583 +⌛️ [2/4] FRONTEND: Frontend time: 7.643s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.580s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14553812 6.03117193 + layer.9.1 0.14559401 6.02349717 + layer.19.0 0.04492324 27.03257536 + layer.19.1 0.04213941 26.97859878 + layer.29.0 4.25320263 287.79251300 + layer.29.1 4.25391672 289.74520637 + layer.39.0 8.72311137 17656.62008450 + layer.39.1 8.87262096 17895.64770881 + ------------------------------------------------------------------------------------- + TOTAL 3.31013081 4524.48391949 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 656644 +BPFP 0.0521 bits/point +EBPFP 0.0521 equivalent bits/point +MSE 4524.483919 +---------------------- --------------------------------------------------------- +Time: 21.363s Load: 1.140s, Pack+Encode: 7.643s, Decode+Unpack: 12.580s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4524.4839 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00471-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst (94/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.162s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 59,124B, BPFP=0.0375 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,788B, BPFP=0.0373 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,116B, BPFP=0.0496 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,152B, BPFP=0.0496 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 96,852B, BPFP=0.0615 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 96,084B, BPFP=0.0610 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,592B, BPFP=0.0594 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,580B, BPFP=0.0588 +⌛️ [2/4] FRONTEND: Frontend time: 7.629s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.777s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14518122 6.07498870 + layer.9.1 0.14529820 6.05508739 + layer.19.0 0.11833418 27.47165461 + layer.19.1 0.12038008 27.17908068 + layer.29.0 4.31360161 308.89259019 + layer.29.1 4.31792870 308.97676308 + layer.39.0 9.40764201 18392.94377641 + layer.39.1 11.30764416 18557.25446864 + ------------------------------------------------------------------------------------- + TOTAL 3.73450127 4704.35605121 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 653288 +BPFP 0.0518 bits/point +EBPFP 0.0518 equivalent bits/point +MSE 4704.356051 +---------------------- --------------------------------------------------------- +Time: 21.568s Load: 1.162s, Pack+Encode: 7.629s, Decode+Unpack: 12.777s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4704.3561 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00472-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst (95/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.158s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,388B, BPFP=0.0371 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,868B, BPFP=0.0374 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,508B, BPFP=0.0492 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,432B, BPFP=0.0498 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 94,916B, BPFP=0.0602 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,860B, BPFP=0.0608 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,328B, BPFP=0.0592 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,156B, BPFP=0.0598 +⌛️ [2/4] FRONTEND: Frontend time: 7.635s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.333s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00315138 6.20596868 + layer.9.1 0.00505826 6.22285366 + layer.19.0 0.09147678 27.83385044 + layer.19.1 0.09143778 27.77480094 + layer.29.0 0.11015094 287.33120734 + layer.29.1 0.11338039 289.02228226 + layer.39.0 9.14784464 18449.70555736 + layer.39.1 8.98944348 18547.35521612 + ------------------------------------------------------------------------------------- + TOTAL 2.31899296 4705.18146710 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 651456 +BPFP 0.0517 bits/point +EBPFP 0.0517 equivalent bits/point +MSE 4705.181467 +---------------------- --------------------------------------------------------- +Time: 21.125s Load: 1.158s, Pack+Encode: 7.635s, Decode+Unpack: 12.333s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4705.1815 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00475-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst (96/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.163s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,568B, BPFP=0.0372 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,900B, BPFP=0.0374 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,592B, BPFP=0.0493 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,324B, BPFP=0.0497 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 94,116B, BPFP=0.0597 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,472B, BPFP=0.0606 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 94,476B, BPFP=0.0600 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 95,828B, BPFP=0.0608 +⌛️ [2/4] FRONTEND: Frontend time: 7.690s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.366s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14591913 6.25661155 + layer.9.1 0.03347605 6.24524763 + layer.19.0 0.12173996 28.14753869 + layer.19.1 0.12099332 28.01433773 + layer.29.0 0.11078974 283.69085148 + layer.29.1 0.11776269 283.12666558 + layer.39.0 10.17800795 19317.31426714 + layer.39.1 9.88744998 19065.69645759 + ------------------------------------------------------------------------------------- + TOTAL 2.58951735 4877.31149717 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 653276 +BPFP 0.0518 bits/point +EBPFP 0.0518 equivalent bits/point +MSE 4877.311497 +---------------------- --------------------------------------------------------- +Time: 21.219s Load: 1.163s, Pack+Encode: 7.690s, Decode+Unpack: 12.366s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4877.3115 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00476-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst (97/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.168s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,464B, BPFP=0.0365 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,376B, BPFP=0.0364 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,848B, BPFP=0.0500 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,432B, BPFP=0.0498 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 100,476B, BPFP=0.0638 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 99,636B, BPFP=0.0632 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 98,216B, BPFP=0.0623 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 97,448B, BPFP=0.0619 +⌛️ [2/4] FRONTEND: Frontend time: 7.519s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.115s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66405266 6.02094801 + layer.9.1 2.66543197 6.01545933 + layer.19.0 3.22131407 25.83236513 + layer.19.1 3.22426883 25.97023024 + layer.29.0 4.27224607 282.22980988 + layer.29.1 4.27784520 284.63188577 + layer.39.0 8.94937744 18316.45888853 + layer.39.1 8.82170070 18497.64835879 + ------------------------------------------------------------------------------------- + TOTAL 4.76202962 4680.60099321 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 667896 +BPFP 0.0530 bits/point +EBPFP 0.0530 equivalent bits/point +MSE 4680.600993 +---------------------- --------------------------------------------------------- +Time: 20.802s Load: 1.168s, Pack+Encode: 7.519s, Decode+Unpack: 12.115s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4680.6010 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00643-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst (98/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.145s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 57,328B, BPFP=0.0364 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 57,248B, BPFP=0.0363 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 76,692B, BPFP=0.0487 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 76,036B, BPFP=0.0483 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 94,316B, BPFP=0.0599 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 95,036B, BPFP=0.0603 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 92,492B, BPFP=0.0587 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,180B, BPFP=0.0598 +⌛️ [2/4] FRONTEND: Frontend time: 7.784s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.100s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00104424 6.07718367 + layer.9.1 0.00091568 6.08719557 + layer.19.0 0.08171424 26.38678603 + layer.19.1 0.08373584 26.41673607 + layer.29.0 4.26071267 284.53042736 + layer.29.1 4.26438533 283.02006825 + layer.39.0 8.39843369 18693.77185570 + layer.39.1 8.51949380 18431.11212220 + ------------------------------------------------------------------------------------- + TOTAL 3.20130444 4719.67529686 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 643328 +BPFP 0.0510 bits/point +EBPFP 0.0510 equivalent bits/point +MSE 4719.675297 +---------------------- --------------------------------------------------------- +Time: 21.029s Load: 1.145s, Pack+Encode: 7.784s, Decode+Unpack: 12.100s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4719.6753 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00644-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst (99/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.147s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,964B, BPFP=0.0374 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 59,256B, BPFP=0.0376 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 78,604B, BPFP=0.0499 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,980B, BPFP=0.0501 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 96,000B, BPFP=0.0609 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 96,780B, BPFP=0.0614 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 91,640B, BPFP=0.0582 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 92,548B, BPFP=0.0587 +⌛️ [2/4] FRONTEND: Frontend time: 7.570s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.127s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14622840 6.16883074 + layer.9.1 0.03344178 6.16514284 + layer.19.0 0.12675888 27.99203516 + layer.19.1 0.12382618 27.97342684 + layer.29.0 0.12223263 297.51074504 + layer.29.1 0.12797405 299.38558255 + layer.39.0 10.69978368 19083.85830354 + layer.39.1 8.63538768 19114.32434189 + ------------------------------------------------------------------------------------- + TOTAL 2.50195416 4857.92230108 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 652772 +BPFP 0.0518 bits/point +EBPFP 0.0518 equivalent bits/point +MSE 4857.922301 +---------------------- --------------------------------------------------------- +Time: 20.844s Load: 1.147s, Pack+Encode: 7.570s, Decode+Unpack: 12.127s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4857.9223 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00272-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst (100/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.161s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 58,644B, BPFP=0.0372 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 58,484B, BPFP=0.0371 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 77,932B, BPFP=0.0495 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 78,748B, BPFP=0.0500 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 96,396B, BPFP=0.0612 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 96,744B, BPFP=0.0614 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 93,984B, BPFP=0.0597 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 94,208B, BPFP=0.0598 +⌛️ [2/4] FRONTEND: Frontend time: 7.666s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.334s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14492994 6.10799886 + layer.9.1 0.14498602 6.11203778 + layer.19.0 0.12957112 27.73447148 + layer.19.1 0.13054295 27.93895992 + layer.29.0 0.16610158 297.40162902 + layer.29.1 0.14872770 296.71581492 + layer.39.0 16.52878844 18921.88235294 + layer.39.1 24.55764797 18893.44296393 + ------------------------------------------------------------------------------------- + TOTAL 5.24391197 4809.66702861 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 655140 +BPFP 0.0520 bits/point +EBPFP 0.0520 equivalent bits/point +MSE 4809.667029 +---------------------- --------------------------------------------------------- +Time: 21.161s Load: 1.161s, Pack+Encode: 7.666s, Decode+Unpack: 12.334s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 4809.6670 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst + to output-fixed/dinov3-total/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00278-wflip.zst +------------------------ ---------------------------- +TOTAL PROCESSING SUMMARY +------------------------ ---------------------------- +Total files 100 +Avg BPFP 0.0522 bits/point +Avg EBPFP 0.0522 equivalent bits/point +Avg MSE 4767.951133 +Avg Time 21.038s +------------------------ ---------------------------- diff --git a/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00350-wflip.zst b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00350-wflip.zst new file mode 100644 index 0000000000000000000000000000000000000000..1d26dac2dc096ede95eedffd5dba92e8dab28cd0 --- /dev/null +++ b/lambda0.001/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00350-wflip.zst @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:de6edb40ced3cc84952f321aa6ac341c08a307d636a03a579cc87927dcef7f54 +size 89278737 diff --git a/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log b/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log new file mode 100644 index 0000000000000000000000000000000000000000..2db0127d0694f468202c8e34fe8197a1a9649f34 --- /dev/null +++ b/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log @@ -0,0 +1,15744 @@ +Experiment: dtufc_hyperprior-featurecoding_dinov3-total_individual +Log file: output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log +DTUFCCodecConfig: + arch: hyperprior-featurecoding + handler: dinov3-total + checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.004_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar + transform_type: kmeans + transform_mapping:featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json + bit_depth: 8 + device: cuda:0 +Loading checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.004_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar +Checkpoint epoch: 559 +Loaded hyperprior-featurecoding (1-channel) on cuda:0 +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_9_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.9' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_9_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_19_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.19' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_19_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_29_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.29' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_29_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_39_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.39' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_39_0.json +Loaded per-key mappings: model=dinov3-total + Keys: ['layer.9', 'layer.19', 'layer.29', 'layer.39'] +---------------- ------------------------------------------------------------------------------------------------------------------------------ +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +Checkpoint codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.004_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar +Transform type kmeans +Transform config featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json +Input ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features +Output output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test +---------------- ------------------------------------------------------------------------------------------------------------------------------ +Files found: 100 +---------------------------------------------------------------------- + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst (1/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.264s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 285,412B, BPFP=0.1812 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 295,556B, BPFP=0.1876 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 489,632B, BPFP=0.3108 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 494,952B, BPFP=0.3142 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 717,472B, BPFP=0.4554 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 708,976B, BPFP=0.4500 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 540,044B, BPFP=0.3428 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 543,896B, BPFP=0.3452 +⌛️ [2/4] FRONTEND: Frontend time: 7.868s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.796s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14561056 4.76883144 + layer.9.1 0.14522085 4.72431136 + layer.19.0 3.25142184 9.02151738 + layer.19.1 3.25206135 8.52946064 + layer.29.0 4.23946030 57.87040441 + layer.29.1 4.24539299 44.28122969 + layer.39.0 32.17105490 2453.25788105 + layer.39.1 19.15684032 2365.69743256 + ------------------------------------------------------------------------------------- + TOTAL 8.32588289 618.51888357 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4075940 +BPFP 0.3234 bits/point +EBPFP 0.3234 equivalent bits/point +MSE 618.518884 +---------------------- --------------------------------------------------------- +Time: 21.928s Load: 1.264s, Pack+Encode: 7.868s, Decode+Unpack: 12.796s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 618.5189 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00512-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst (2/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.283s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 320,228B, BPFP=0.2033 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 295,668B, BPFP=0.1877 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 680,040B, BPFP=0.4317 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 654,000B, BPFP=0.4151 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 809,844B, BPFP=0.5140 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 782,292B, BPFP=0.4966 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 563,004B, BPFP=0.3574 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 556,708B, BPFP=0.3534 +⌛️ [2/4] FRONTEND: Frontend time: 7.671s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.328s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03307773 4.66625937 + layer.9.1 0.03291117 4.68239312 + layer.19.0 0.04156009 16.87722035 + layer.19.1 0.03760627 7.72617467 + layer.29.0 4.28582750 83.91094715 + layer.29.1 4.28551552 73.64790177 + layer.39.0 9.83402183 2839.41046474 + layer.39.1 9.85397836 2798.85472863 + ------------------------------------------------------------------------------------- + TOTAL 3.55056231 728.72201123 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4661784 +BPFP 0.3699 bits/point +EBPFP 0.3699 equivalent bits/point +MSE 728.722011 +---------------------- --------------------------------------------------------- +Time: 21.282s Load: 1.283s, Pack+Encode: 7.671s, Decode+Unpack: 12.328s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 728.7220 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00657-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst (3/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.257s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 325,140B, BPFP=0.2064 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 321,012B, BPFP=0.2038 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 735,056B, BPFP=0.4666 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 741,140B, BPFP=0.4704 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 831,396B, BPFP=0.5277 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 834,776B, BPFP=0.5299 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 552,800B, BPFP=0.3509 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 558,240B, BPFP=0.3543 +⌛️ [2/4] FRONTEND: Frontend time: 7.695s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.396s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00083900 4.68476328 + layer.9.1 0.00259629 4.67373291 + layer.19.0 0.00955961 12.26618358 + layer.19.1 0.08538111 21.36145444 + layer.29.0 0.11631418 120.90951007 + layer.29.1 0.11200302 105.44379672 + layer.39.0 14.47657393 3126.90965226 + layer.39.1 13.08093694 2989.99837504 + ------------------------------------------------------------------------------------- + TOTAL 3.48552551 798.28093354 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4899560 +BPFP 0.3887 bits/point +EBPFP 0.3887 equivalent bits/point +MSE 798.280934 +---------------------- --------------------------------------------------------- +Time: 21.348s Load: 1.257s, Pack+Encode: 7.695s, Decode+Unpack: 12.396s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 798.2809 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00676-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst (4/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.253s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 295,700B, BPFP=0.1877 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 294,256B, BPFP=0.1868 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 606,848B, BPFP=0.3852 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 610,404B, BPFP=0.3875 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 827,100B, BPFP=0.5250 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 773,396B, BPFP=0.4909 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 553,068B, BPFP=0.3511 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 539,840B, BPFP=0.3427 +⌛️ [2/4] FRONTEND: Frontend time: 7.690s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.375s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.70169554 4.80084726 + layer.9.1 0.03294074 4.75218037 + layer.19.0 3.25671692 8.22463286 + layer.19.1 3.25834093 8.53096182 + layer.29.0 0.10810242 145.12722416 + layer.29.1 0.10661203 128.15783434 + layer.39.0 8.95005916 2725.98440039 + layer.39.1 8.98756017 2702.15940851 + ------------------------------------------------------------------------------------- + TOTAL 3.42525349 715.96718621 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4500612 +BPFP 0.3571 bits/point +EBPFP 0.3571 equivalent bits/point +MSE 715.967186 +---------------------- --------------------------------------------------------- +Time: 21.318s Load: 1.253s, Pack+Encode: 7.690s, Decode+Unpack: 12.375s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 715.9672 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00688-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst (5/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.261s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 326,300B, BPFP=0.2071 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 320,112B, BPFP=0.2032 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 644,672B, BPFP=0.4092 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 657,504B, BPFP=0.4174 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 786,888B, BPFP=0.4995 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 805,180B, BPFP=0.5111 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 543,552B, BPFP=0.3450 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 560,756B, BPFP=0.3559 +⌛️ [2/4] FRONTEND: Frontend time: 7.669s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.279s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14527870 4.69414361 + layer.9.1 0.14521496 4.62768880 + layer.19.0 0.03964342 21.57238686 + layer.19.1 0.03956446 8.07167277 + layer.29.0 0.12258449 94.94081085 + layer.29.1 0.12735008 131.17522140 + layer.39.0 32.94776263 2951.67500812 + layer.39.1 29.25669534 2981.24634384 + ------------------------------------------------------------------------------------- + TOTAL 7.85301176 774.75040953 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4644964 +BPFP 0.3685 bits/point +EBPFP 0.3685 equivalent bits/point +MSE 774.750410 +---------------------- --------------------------------------------------------- +Time: 21.208s Load: 1.261s, Pack+Encode: 7.669s, Decode+Unpack: 12.279s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 774.7504 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00693-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst (6/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.264s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 302,936B, BPFP=0.1923 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 292,588B, BPFP=0.1857 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 539,932B, BPFP=0.3427 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 584,728B, BPFP=0.3712 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 760,176B, BPFP=0.4825 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 796,228B, BPFP=0.5054 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 524,392B, BPFP=0.3329 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 535,828B, BPFP=0.3401 +⌛️ [2/4] FRONTEND: Frontend time: 7.711s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.306s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66816044 4.77541252 + layer.9.1 2.66817504 4.79760084 + layer.19.0 3.22262959 9.70201609 + layer.19.1 3.22037432 8.46666929 + layer.29.0 4.30448692 117.22166477 + layer.29.1 4.31085282 150.43438211 + layer.39.0 38.33931691 2693.75609360 + layer.39.1 57.25219370 2783.77088073 + ------------------------------------------------------------------------------------- + TOTAL 14.49827372 721.61558999 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4336808 +BPFP 0.3441 bits/point +EBPFP 0.3441 equivalent bits/point +MSE 721.615590 +---------------------- --------------------------------------------------------- +Time: 21.282s Load: 1.264s, Pack+Encode: 7.711s, Decode+Unpack: 12.306s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 721.6156 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00731-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst (7/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.258s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 301,208B, BPFP=0.1912 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 307,456B, BPFP=0.1952 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 648,552B, BPFP=0.4117 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 568,708B, BPFP=0.3610 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 734,920B, BPFP=0.4665 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 721,768B, BPFP=0.4581 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 524,484B, BPFP=0.3329 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 505,660B, BPFP=0.3210 +⌛️ [2/4] FRONTEND: Frontend time: 7.672s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.256s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11332920 4.74381119 + layer.9.1 0.00092169 4.73862719 + layer.19.0 3.23006092 8.32895208 + layer.19.1 3.23257961 22.00066522 + layer.29.0 4.28548854 144.05425333 + layer.29.1 4.27808990 118.99681102 + layer.39.0 10.57841825 2508.41777706 + layer.39.1 20.33118703 2478.87162821 + ------------------------------------------------------------------------------------- + TOTAL 5.75625939 661.26906566 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4312756 +BPFP 0.3422 bits/point +EBPFP 0.3422 equivalent bits/point +MSE 661.269066 +---------------------- --------------------------------------------------------- +Time: 21.186s Load: 1.258s, Pack+Encode: 7.672s, Decode+Unpack: 12.256s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 661.2691 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00061-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst (8/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.251s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 310,232B, BPFP=0.1969 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 316,852B, BPFP=0.2011 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 645,280B, BPFP=0.4096 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 609,948B, BPFP=0.3872 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 790,032B, BPFP=0.5015 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 744,760B, BPFP=0.4727 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 499,724B, BPFP=0.3172 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 495,040B, BPFP=0.3142 +⌛️ [2/4] FRONTEND: Frontend time: 7.699s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.265s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14443954 4.64575346 + layer.9.1 0.14435121 4.61312351 + layer.19.0 0.03807715 8.47181584 + layer.19.1 0.03781311 8.18717183 + layer.29.0 0.10781899 111.92645028 + layer.29.1 0.10618912 53.66602515 + layer.39.0 9.30898666 2730.62755931 + layer.39.1 9.83625107 2911.16542086 + ------------------------------------------------------------------------------------- + TOTAL 2.46549086 729.16291503 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4411868 +BPFP 0.3501 bits/point +EBPFP 0.3501 equivalent bits/point +MSE 729.162915 +---------------------- --------------------------------------------------------- +Time: 21.215s Load: 1.251s, Pack+Encode: 7.699s, Decode+Unpack: 12.265s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 729.1629 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00175-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst (9/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.264s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 312,280B, BPFP=0.1982 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 318,880B, BPFP=0.2024 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 759,356B, BPFP=0.4820 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 721,868B, BPFP=0.4582 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 842,380B, BPFP=0.5347 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 832,456B, BPFP=0.5284 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 551,224B, BPFP=0.3499 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 545,008B, BPFP=0.3459 +⌛️ [2/4] FRONTEND: Frontend time: 7.723s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.285s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14543928 4.63651753 + layer.9.1 0.14562574 4.66298374 + layer.19.0 0.11552505 7.93162856 + layer.19.1 0.12052174 21.62991042 + layer.29.0 0.10841144 62.06061606 + layer.29.1 0.10845811 48.77129712 + layer.39.0 9.17501701 3129.54631134 + layer.39.1 9.20635778 2915.88592785 + ------------------------------------------------------------------------------------- + TOTAL 2.39066952 774.39064908 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4883452 +BPFP 0.3875 bits/point +EBPFP 0.3875 equivalent bits/point +MSE 774.390649 +---------------------- --------------------------------------------------------- +Time: 21.272s Load: 1.264s, Pack+Encode: 7.723s, Decode+Unpack: 12.285s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 774.3906 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00186-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst (10/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.261s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 296,872B, BPFP=0.1884 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 294,748B, BPFP=0.1871 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 502,912B, BPFP=0.3192 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 536,564B, BPFP=0.3406 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 653,928B, BPFP=0.4151 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 663,736B, BPFP=0.4213 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 501,568B, BPFP=0.3184 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 508,108B, BPFP=0.3225 +⌛️ [2/4] FRONTEND: Frontend time: 7.679s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.286s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.78226592 4.72658694 + layer.9.1 2.78427046 4.66927538 + layer.19.0 3.22580366 9.00983672 + layer.19.1 3.22969594 8.67494592 + layer.29.0 4.29525448 80.92864397 + layer.29.1 0.11349234 115.99516575 + layer.39.0 8.89338553 2620.13990900 + layer.39.1 8.88767087 2558.59798505 + ------------------------------------------------------------------------------------- + TOTAL 4.27647990 675.34279359 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 3958436 +BPFP 0.3141 bits/point +EBPFP 0.3141 equivalent bits/point +MSE 675.342794 +---------------------- --------------------------------------------------------- +Time: 21.226s Load: 1.261s, Pack+Encode: 7.679s, Decode+Unpack: 12.286s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 675.3428 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00282-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst (11/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.259s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 293,664B, BPFP=0.1864 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 298,724B, BPFP=0.1896 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 537,896B, BPFP=0.3414 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 594,436B, BPFP=0.3773 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 680,448B, BPFP=0.4319 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 683,548B, BPFP=0.4339 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 482,680B, BPFP=0.3064 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 493,544B, BPFP=0.3133 +⌛️ [2/4] FRONTEND: Frontend time: 7.719s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.309s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14514284 4.65863793 + layer.9.1 0.14518188 4.68219730 + layer.19.0 0.04057091 49.69918955 + layer.19.1 0.04041447 36.65774801 + layer.29.0 4.25641542 50.13694345 + layer.29.1 4.26613502 60.74415522 + layer.39.0 12.58558458 2914.08514787 + layer.39.1 8.96866240 2937.11309717 + ------------------------------------------------------------------------------------- + TOTAL 3.80601344 757.22213956 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4064940 +BPFP 0.3225 bits/point +EBPFP 0.3225 equivalent bits/point +MSE 757.222140 +---------------------- --------------------------------------------------------- +Time: 21.287s Load: 1.259s, Pack+Encode: 7.719s, Decode+Unpack: 12.309s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 757.2221 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00524-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst (12/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.265s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 301,924B, BPFP=0.1916 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 302,304B, BPFP=0.1919 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 595,984B, BPFP=0.3783 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 561,372B, BPFP=0.3563 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 715,348B, BPFP=0.4541 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 703,120B, BPFP=0.4463 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 501,496B, BPFP=0.3183 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 518,196B, BPFP=0.3289 +⌛️ [2/4] FRONTEND: Frontend time: 7.657s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.274s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67015388 4.73128440 + layer.9.1 0.00076871 4.68866889 + layer.19.0 3.22151687 7.75921974 + layer.19.1 3.22388957 7.74519241 + layer.29.0 4.24084786 57.61570219 + layer.29.1 4.24602234 49.18678400 + layer.39.0 7.87160790 2428.97790055 + layer.39.1 9.85764150 2529.90087748 + ------------------------------------------------------------------------------------- + TOTAL 4.41655608 636.32570371 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4199744 +BPFP 0.3332 bits/point +EBPFP 0.3332 equivalent bits/point +MSE 636.325704 +---------------------- --------------------------------------------------------- +Time: 21.196s Load: 1.265s, Pack+Encode: 7.657s, Decode+Unpack: 12.274s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 636.3257 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00538-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst (13/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.252s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 316,592B, BPFP=0.2010 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 312,912B, BPFP=0.1986 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 700,764B, BPFP=0.4448 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 724,404B, BPFP=0.4598 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 822,860B, BPFP=0.5223 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 809,952B, BPFP=0.5141 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 542,208B, BPFP=0.3442 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 546,948B, BPFP=0.3472 +⌛️ [2/4] FRONTEND: Frontend time: 7.771s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.264s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00068237 4.66002676 + layer.9.1 0.00070576 4.66083226 + layer.19.0 0.00823322 26.50716506 + layer.19.1 0.08594799 36.04897729 + layer.29.0 0.12200666 134.06700926 + layer.29.1 0.12451052 184.19832629 + layer.39.0 55.99513528 3289.23171921 + layer.39.1 28.81185256 3295.24569386 + ------------------------------------------------------------------------------------- + TOTAL 10.64363429 871.82746875 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4776640 +BPFP 0.3790 bits/point +EBPFP 0.3790 equivalent bits/point +MSE 871.827469 +---------------------- --------------------------------------------------------- +Time: 21.287s Load: 1.252s, Pack+Encode: 7.771s, Decode+Unpack: 12.264s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 871.8275 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00927-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst (14/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.260s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 321,964B, BPFP=0.2044 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 311,856B, BPFP=0.1980 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 615,696B, BPFP=0.3908 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 605,052B, BPFP=0.3841 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 713,236B, BPFP=0.4527 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 722,956B, BPFP=0.4589 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 519,320B, BPFP=0.3296 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 532,396B, BPFP=0.3379 +⌛️ [2/4] FRONTEND: Frontend time: 7.681s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.298s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14556937 4.64406630 + layer.9.1 0.03327741 4.64716800 + layer.19.0 0.11590617 36.08881926 + layer.19.1 0.11733878 22.14747268 + layer.29.0 0.11334742 113.03111797 + layer.29.1 4.29039579 82.28161562 + layer.39.0 9.10722066 2668.30224244 + layer.39.1 44.52401893 2660.14592135 + ------------------------------------------------------------------------------------- + TOTAL 7.30588432 698.91105295 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4342476 +BPFP 0.3445 bits/point +EBPFP 0.3445 equivalent bits/point +MSE 698.911053 +---------------------- --------------------------------------------------------- +Time: 21.240s Load: 1.260s, Pack+Encode: 7.681s, Decode+Unpack: 12.298s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 698.9111 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00933-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst (15/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.254s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 301,180B, BPFP=0.1912 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 303,444B, BPFP=0.1926 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 641,008B, BPFP=0.4069 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 631,068B, BPFP=0.4006 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 733,736B, BPFP=0.4657 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 757,880B, BPFP=0.4811 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 478,900B, BPFP=0.3040 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 484,220B, BPFP=0.3074 +⌛️ [2/4] FRONTEND: Frontend time: 7.682s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.283s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11318631 4.67186516 + layer.9.1 0.11319129 4.68783451 + layer.19.0 0.00665199 7.96597995 + layer.19.1 0.00853768 8.34821610 + layer.29.0 4.27225940 54.44515254 + layer.29.1 4.27324961 55.89158982 + layer.39.0 14.80262837 2308.52518687 + layer.39.1 16.56649765 2363.36529087 + ------------------------------------------------------------------------------------- + TOTAL 5.01952529 600.98763948 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4331436 +BPFP 0.3437 bits/point +EBPFP 0.3437 equivalent bits/point +MSE 600.987639 +---------------------- --------------------------------------------------------- +Time: 21.219s Load: 1.254s, Pack+Encode: 7.682s, Decode+Unpack: 12.283s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 600.9876 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01095-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst (16/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.261s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 314,536B, BPFP=0.1997 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 304,716B, BPFP=0.1934 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 633,872B, BPFP=0.4024 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 643,176B, BPFP=0.4083 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 759,472B, BPFP=0.4821 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 749,920B, BPFP=0.4760 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 522,484B, BPFP=0.3316 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 514,772B, BPFP=0.3268 +⌛️ [2/4] FRONTEND: Frontend time: 7.726s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.333s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00065117 4.70930238 + layer.9.1 0.00066201 4.69464062 + layer.19.0 0.00984582 8.13014084 + layer.19.1 0.01156107 8.11496397 + layer.29.0 4.26547583 87.39353469 + layer.29.1 4.26296603 132.46489072 + layer.39.0 11.21169412 2705.50048749 + layer.39.1 9.31977106 2593.79168021 + ------------------------------------------------------------------------------------- + TOTAL 3.63532839 693.09995512 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4442948 +BPFP 0.3525 bits/point +EBPFP 0.3525 equivalent bits/point +MSE 693.099955 +---------------------- --------------------------------------------------------- +Time: 21.320s Load: 1.261s, Pack+Encode: 7.726s, Decode+Unpack: 12.333s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 693.1000 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01125-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst (17/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.262s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 287,484B, BPFP=0.1825 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 290,932B, BPFP=0.1847 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 570,472B, BPFP=0.3621 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 548,056B, BPFP=0.3479 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 687,152B, BPFP=0.4362 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 686,156B, BPFP=0.4355 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 497,148B, BPFP=0.3156 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 480,296B, BPFP=0.3049 +⌛️ [2/4] FRONTEND: Frontend time: 7.670s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.249s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00085355 4.74313296 + layer.9.1 0.00085581 4.74771141 + layer.19.0 0.00808159 8.72730008 + layer.19.1 0.00635426 9.12152475 + layer.29.0 4.24551200 59.41871649 + layer.29.1 4.24803037 64.81056020 + layer.39.0 9.19283951 2396.35391615 + layer.39.1 9.46657027 2318.46928827 + ------------------------------------------------------------------------------------- + TOTAL 3.39613717 608.29901879 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4047696 +BPFP 0.3212 bits/point +EBPFP 0.3212 equivalent bits/point +MSE 608.299019 +---------------------- --------------------------------------------------------- +Time: 21.181s Load: 1.262s, Pack+Encode: 7.670s, Decode+Unpack: 12.249s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 608.2990 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01135-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst (18/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.256s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 320,968B, BPFP=0.2037 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 309,596B, BPFP=0.1965 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 622,596B, BPFP=0.3952 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 623,508B, BPFP=0.3958 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 770,580B, BPFP=0.4891 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 786,912B, BPFP=0.4995 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 523,324B, BPFP=0.3322 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 530,940B, BPFP=0.3370 +⌛️ [2/4] FRONTEND: Frontend time: 7.709s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.304s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67056234 4.79371807 + layer.9.1 2.67147828 4.76389975 + layer.19.0 0.00618387 7.86398455 + layer.19.1 0.08383032 12.77198138 + layer.29.0 4.28489822 88.69419280 + layer.29.1 4.28470970 60.86809900 + layer.39.0 10.15376305 2901.50341241 + layer.39.1 8.47863686 2823.12642184 + ------------------------------------------------------------------------------------- + TOTAL 4.07925783 738.04821373 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4488424 +BPFP 0.3561 bits/point +EBPFP 0.3561 equivalent bits/point +MSE 738.048214 +---------------------- --------------------------------------------------------- +Time: 21.269s Load: 1.256s, Pack+Encode: 7.709s, Decode+Unpack: 12.304s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 738.0482 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01161-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst (19/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.256s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 305,744B, BPFP=0.1941 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 317,264B, BPFP=0.2014 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 634,012B, BPFP=0.4024 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 650,344B, BPFP=0.4128 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 780,808B, BPFP=0.4956 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 782,460B, BPFP=0.4967 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 522,844B, BPFP=0.3319 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 517,040B, BPFP=0.3282 +⌛️ [2/4] FRONTEND: Frontend time: 7.717s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.246s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67132404 4.76220782 + layer.9.1 2.67117709 4.77122000 + layer.19.0 0.00597838 12.37053136 + layer.19.1 0.00605309 7.77983006 + layer.29.0 4.29273040 150.01796596 + layer.29.1 4.29206328 69.55420255 + layer.39.0 9.96127074 2629.93678908 + layer.39.1 10.21295854 2524.82287943 + ------------------------------------------------------------------------------------- + TOTAL 4.26419445 675.50195328 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4510516 +BPFP 0.3579 bits/point +EBPFP 0.3579 equivalent bits/point +MSE 675.501953 +---------------------- --------------------------------------------------------- +Time: 21.220s Load: 1.256s, Pack+Encode: 7.717s, Decode+Unpack: 12.246s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 675.5020 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01162-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst (20/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.266s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 315,992B, BPFP=0.2006 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 304,720B, BPFP=0.1934 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 611,572B, BPFP=0.3882 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 610,528B, BPFP=0.3875 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 706,232B, BPFP=0.4483 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 691,364B, BPFP=0.4388 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 492,360B, BPFP=0.3125 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 485,612B, BPFP=0.3082 +⌛️ [2/4] FRONTEND: Frontend time: 7.672s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.263s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14568686 4.73144087 + layer.9.1 0.14558674 4.69772423 + layer.19.0 0.00960369 8.03508897 + layer.19.1 0.03847206 22.66249848 + layer.29.0 4.24438723 58.48910769 + layer.29.1 4.24578970 102.14787130 + layer.39.0 9.23757985 2637.39129022 + layer.39.1 9.43674592 2738.46798830 + ------------------------------------------------------------------------------------- + TOTAL 3.43798151 697.07787626 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4218380 +BPFP 0.3347 bits/point +EBPFP 0.3347 equivalent bits/point +MSE 697.077876 +---------------------- --------------------------------------------------------- +Time: 21.201s Load: 1.266s, Pack+Encode: 7.672s, Decode+Unpack: 12.263s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 697.0779 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01163-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst (21/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.265s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 327,984B, BPFP=0.2082 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 317,216B, BPFP=0.2014 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 798,264B, BPFP=0.5067 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 725,952B, BPFP=0.4608 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 828,956B, BPFP=0.5262 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 809,288B, BPFP=0.5137 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 537,356B, BPFP=0.3411 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 533,552B, BPFP=0.3387 +⌛️ [2/4] FRONTEND: Frontend time: 7.711s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.276s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11307967 4.69941521 + layer.9.1 0.00073224 4.66209858 + layer.19.0 0.08207503 7.83069955 + layer.19.1 0.08214869 7.28589002 + layer.29.0 4.26728487 76.67411440 + layer.29.1 4.26774951 54.51577734 + layer.39.0 12.81553410 2870.86805330 + layer.39.1 23.05196315 2871.49366266 + ------------------------------------------------------------------------------------- + TOTAL 5.58507091 737.25371388 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4878568 +BPFP 0.3871 bits/point +EBPFP 0.3871 equivalent bits/point +MSE 737.253714 +---------------------- --------------------------------------------------------- +Time: 21.252s Load: 1.265s, Pack+Encode: 7.711s, Decode+Unpack: 12.276s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 737.2537 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01195-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst (22/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.251s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 331,184B, BPFP=0.2102 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 331,916B, BPFP=0.2107 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 697,080B, BPFP=0.4425 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 745,636B, BPFP=0.4733 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 953,984B, BPFP=0.6055 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 960,676B, BPFP=0.6098 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 493,060B, BPFP=0.3130 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 497,400B, BPFP=0.3157 +⌛️ [2/4] FRONTEND: Frontend time: 7.734s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.317s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14519844 4.61045185 + layer.9.1 0.14499054 4.57308572 + layer.19.0 0.12156012 26.26594237 + layer.19.1 0.12030756 49.67093049 + layer.29.0 0.12020218 112.24337829 + layer.29.1 0.12115470 153.54677852 + layer.39.0 8.85439666 2936.41143971 + layer.39.1 8.75438231 2813.11244719 + ------------------------------------------------------------------------------------- + TOTAL 2.29777406 762.55430677 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5010936 +BPFP 0.3976 bits/point +EBPFP 0.3976 equivalent bits/point +MSE 762.554307 +---------------------- --------------------------------------------------------- +Time: 21.302s Load: 1.251s, Pack+Encode: 7.734s, Decode+Unpack: 12.317s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 762.5543 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00083-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst (23/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.264s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 332,216B, BPFP=0.2109 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 331,012B, BPFP=0.2101 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 721,192B, BPFP=0.4578 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 676,972B, BPFP=0.4297 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 911,088B, BPFP=0.5783 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 893,984B, BPFP=0.5675 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 482,104B, BPFP=0.3060 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 476,476B, BPFP=0.3024 +⌛️ [2/4] FRONTEND: Frontend time: 7.723s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.285s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14541288 21.00844471 + layer.9.1 0.14479464 4.56791124 + layer.19.0 0.11855170 26.47988097 + layer.19.1 0.11778439 36.09712687 + layer.29.0 0.12648388 121.87221726 + layer.29.1 0.12520221 162.72727088 + layer.39.0 8.37129624 2737.83035424 + layer.39.1 8.45478741 2794.99805005 + ------------------------------------------------------------------------------------- + TOTAL 2.20053917 738.19765703 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4825044 +BPFP 0.3828 bits/point +EBPFP 0.3828 equivalent bits/point +MSE 738.197657 +---------------------- --------------------------------------------------------- +Time: 21.272s Load: 1.264s, Pack+Encode: 7.723s, Decode+Unpack: 12.285s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 738.1977 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00084-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst (24/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.260s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 338,064B, BPFP=0.2146 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 332,172B, BPFP=0.2108 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 832,116B, BPFP=0.5282 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 905,000B, BPFP=0.5744 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,039,524B, BPFP=0.6598 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,043,952B, BPFP=0.6626 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 581,376B, BPFP=0.3690 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 595,552B, BPFP=0.3780 +⌛️ [2/4] FRONTEND: Frontend time: 7.758s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.315s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14458197 4.83892368 + layer.9.1 0.14461228 4.58105627 + layer.19.0 0.12127609 45.24306853 + layer.19.1 0.12505172 49.82648988 + layer.29.0 0.11568762 136.38589738 + layer.29.1 0.11796058 135.34205395 + layer.39.0 8.63782956 3062.67143321 + layer.39.1 8.69862780 3148.68703282 + ------------------------------------------------------------------------------------- + TOTAL 2.26320345 823.44699447 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5667756 +BPFP 0.4497 bits/point +EBPFP 0.4497 equivalent bits/point +MSE 823.446994 +---------------------- --------------------------------------------------------- +Time: 21.333s Load: 1.260s, Pack+Encode: 7.758s, Decode+Unpack: 12.315s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 823.4470 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00085-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst (25/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.269s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 337,612B, BPFP=0.2143 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 334,636B, BPFP=0.2124 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 828,224B, BPFP=0.5257 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 830,084B, BPFP=0.5269 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,060,784B, BPFP=0.6733 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,048,036B, BPFP=0.6652 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 619,832B, BPFP=0.3934 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 585,728B, BPFP=0.3718 +⌛️ [2/4] FRONTEND: Frontend time: 7.721s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.328s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14478087 4.57486969 + layer.9.1 0.14472154 4.55330946 + layer.19.0 0.13423899 63.64449951 + layer.19.1 0.13534726 83.18477312 + layer.29.0 0.11251127 144.47636700 + layer.29.1 0.11242151 165.36443776 + layer.39.0 10.58490794 3345.54566136 + layer.39.1 8.80008176 3328.67923302 + ------------------------------------------------------------------------------------- + TOTAL 2.52112639 892.50289386 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5644936 +BPFP 0.4479 bits/point +EBPFP 0.4479 equivalent bits/point +MSE 892.502894 +---------------------- --------------------------------------------------------- +Time: 21.318s Load: 1.269s, Pack+Encode: 7.721s, Decode+Unpack: 12.328s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 892.5029 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00086-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst (26/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.258s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 328,584B, BPFP=0.2086 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 334,144B, BPFP=0.2121 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 734,104B, BPFP=0.4660 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 741,976B, BPFP=0.4710 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 916,464B, BPFP=0.5817 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 917,132B, BPFP=0.5821 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 545,496B, BPFP=0.3463 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 551,520B, BPFP=0.3501 +⌛️ [2/4] FRONTEND: Frontend time: 7.711s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.265s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14590075 4.62124704 + layer.9.1 0.14620647 8.87371209 + layer.19.0 0.11628058 21.53101641 + layer.19.1 0.11601873 35.38386111 + layer.29.0 0.11558260 146.28843029 + layer.29.1 0.11828149 177.34749756 + layer.39.0 28.43028163 3203.85960351 + layer.39.1 24.81181701 2972.27981800 + ------------------------------------------------------------------------------------- + TOTAL 6.75004616 821.27314825 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5069420 +BPFP 0.4022 bits/point +EBPFP 0.4022 equivalent bits/point +MSE 821.273148 +---------------------- --------------------------------------------------------- +Time: 21.234s Load: 1.258s, Pack+Encode: 7.711s, Decode+Unpack: 12.265s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 821.2731 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00090-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst (27/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.264s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 320,680B, BPFP=0.2036 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 331,508B, BPFP=0.2104 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 750,072B, BPFP=0.4761 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 742,976B, BPFP=0.4716 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 934,008B, BPFP=0.5929 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 931,244B, BPFP=0.5911 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 533,484B, BPFP=0.3386 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 527,528B, BPFP=0.3348 +⌛️ [2/4] FRONTEND: Frontend time: 7.744s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.275s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11529700 4.64167900 + layer.9.1 0.14629077 4.65241738 + layer.19.0 0.09721754 21.58186241 + layer.19.1 0.12446257 39.94105206 + layer.29.0 4.28687864 127.60332710 + layer.29.1 4.28715508 87.32876991 + layer.39.0 11.34089363 3193.89632759 + layer.39.1 19.75513766 3106.49756256 + ------------------------------------------------------------------------------------- + TOTAL 5.01916661 823.26787475 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5071500 +BPFP 0.4024 bits/point +EBPFP 0.4024 equivalent bits/point +MSE 823.267875 +---------------------- --------------------------------------------------------- +Time: 21.284s Load: 1.264s, Pack+Encode: 7.744s, Decode+Unpack: 12.275s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 823.2679 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00116-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst (28/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.261s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 344,684B, BPFP=0.2188 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 320,056B, BPFP=0.2032 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 742,768B, BPFP=0.4715 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 728,824B, BPFP=0.4626 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 886,940B, BPFP=0.5630 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 889,908B, BPFP=0.5649 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 600,896B, BPFP=0.3814 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 579,952B, BPFP=0.3681 +⌛️ [2/4] FRONTEND: Frontend time: 7.787s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.295s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14538718 4.64507682 + layer.9.1 0.14538559 4.64145335 + layer.19.0 0.11434236 35.70562084 + layer.19.1 0.11406084 40.51796342 + layer.29.0 0.11219077 123.70509019 + layer.29.1 0.11281304 99.95456207 + layer.39.0 79.88316542 3175.30419240 + layer.39.1 46.71980622 3228.51576211 + ------------------------------------------------------------------------------------- + TOTAL 15.91839393 839.12371515 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5094028 +BPFP 0.4042 bits/point +EBPFP 0.4042 equivalent bits/point +MSE 839.123715 +---------------------- --------------------------------------------------------- +Time: 21.342s Load: 1.261s, Pack+Encode: 7.787s, Decode+Unpack: 12.295s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 839.1237 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00314-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst (29/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.269s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 340,976B, BPFP=0.2164 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 346,404B, BPFP=0.2199 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 770,764B, BPFP=0.4892 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 776,804B, BPFP=0.4931 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 927,992B, BPFP=0.5890 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 919,568B, BPFP=0.5837 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 599,280B, BPFP=0.3804 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 589,992B, BPFP=0.3745 +⌛️ [2/4] FRONTEND: Frontend time: 7.771s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.311s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14512126 4.65407090 + layer.9.1 0.14517278 4.64044187 + layer.19.0 0.11689420 31.22500102 + layer.19.1 0.12099910 87.99621181 + layer.29.0 0.11847120 164.02541030 + layer.29.1 0.12399357 134.81030630 + layer.39.0 75.86630139 3245.90185245 + layer.39.1 56.61936342 3246.78225544 + ------------------------------------------------------------------------------------- + TOTAL 16.65703962 865.00444376 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5271780 +BPFP 0.4183 bits/point +EBPFP 0.4183 equivalent bits/point +MSE 865.004444 +---------------------- --------------------------------------------------------- +Time: 21.351s Load: 1.269s, Pack+Encode: 7.771s, Decode+Unpack: 12.311s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 865.0044 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00316-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst (30/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.265s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 334,032B, BPFP=0.2120 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 336,988B, BPFP=0.2139 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 798,544B, BPFP=0.5069 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 799,816B, BPFP=0.5077 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 895,548B, BPFP=0.5684 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 877,896B, BPFP=0.5572 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 648,788B, BPFP=0.4118 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 604,460B, BPFP=0.3837 +⌛️ [2/4] FRONTEND: Frontend time: 7.740s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.349s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11488649 4.66296438 + layer.9.1 0.14606862 4.65431719 + layer.19.0 0.08767178 26.22253565 + layer.19.1 0.11443626 17.41845751 + layer.29.0 0.10933029 173.74482044 + layer.29.1 0.10817130 152.81753737 + layer.39.0 52.66717785 3277.60155996 + layer.39.1 62.91127214 3258.31751706 + ------------------------------------------------------------------------------------- + TOTAL 14.53237684 864.42996370 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5296072 +BPFP 0.4202 bits/point +EBPFP 0.4202 equivalent bits/point +MSE 864.429964 +---------------------- --------------------------------------------------------- +Time: 21.355s Load: 1.265s, Pack+Encode: 7.740s, Decode+Unpack: 12.349s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 864.4300 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00324-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst (31/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.260s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 337,748B, BPFP=0.2144 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 337,812B, BPFP=0.2144 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 737,736B, BPFP=0.4683 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 732,656B, BPFP=0.4651 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 867,872B, BPFP=0.5509 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 878,244B, BPFP=0.5575 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 568,892B, BPFP=0.3611 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 560,356B, BPFP=0.3557 +⌛️ [2/4] FRONTEND: Frontend time: 7.665s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.280s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14509552 4.64528152 + layer.9.1 0.14520687 4.66399490 + layer.19.0 0.12118574 13.17712946 + layer.19.1 0.11709642 22.48564450 + layer.29.0 0.10963326 134.71045255 + layer.29.1 0.10842036 125.01954014 + layer.39.0 53.79489966 3253.03282418 + layer.39.1 62.27410526 3198.26909327 + ------------------------------------------------------------------------------------- + TOTAL 14.60195539 844.50049507 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5021316 +BPFP 0.3984 bits/point +EBPFP 0.3984 equivalent bits/point +MSE 844.500495 +---------------------- --------------------------------------------------------- +Time: 21.205s Load: 1.260s, Pack+Encode: 7.665s, Decode+Unpack: 12.280s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 844.5005 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00325-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst (32/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.261s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 345,368B, BPFP=0.2192 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 350,676B, BPFP=0.2226 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 824,476B, BPFP=0.5233 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 828,872B, BPFP=0.5261 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 922,772B, BPFP=0.5857 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 919,564B, BPFP=0.5837 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 582,224B, BPFP=0.3696 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 628,244B, BPFP=0.3988 +⌛️ [2/4] FRONTEND: Frontend time: 7.741s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.287s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14577075 4.59010082 + layer.9.1 0.14541274 4.59725794 + layer.19.0 0.13069581 46.05780793 + layer.19.1 0.13545482 63.21810509 + layer.29.0 0.11331055 129.24445483 + layer.29.1 0.11244963 104.63533880 + layer.39.0 32.27446072 3215.59668508 + layer.39.1 16.59366367 3198.13584660 + ------------------------------------------------------------------------------------- + TOTAL 6.20640234 845.75944964 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5402196 +BPFP 0.4286 bits/point +EBPFP 0.4286 equivalent bits/point +MSE 845.759450 +---------------------- --------------------------------------------------------- +Time: 21.289s Load: 1.261s, Pack+Encode: 7.741s, Decode+Unpack: 12.287s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 845.7594 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00327-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst (33/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.263s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 328,628B, BPFP=0.2086 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 321,328B, BPFP=0.2040 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 721,312B, BPFP=0.4579 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 696,600B, BPFP=0.4422 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 851,320B, BPFP=0.5404 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 845,088B, BPFP=0.5364 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 543,460B, BPFP=0.3450 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 571,308B, BPFP=0.3626 +⌛️ [2/4] FRONTEND: Frontend time: 7.698s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.292s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14585920 4.67713517 + layer.9.1 0.14576220 4.66050917 + layer.19.0 0.12270736 36.35736919 + layer.19.1 0.12453605 35.64060215 + layer.29.0 0.11393550 159.04578323 + layer.29.1 0.11678154 149.28822717 + layer.39.0 53.83016636 2960.50666233 + layer.39.1 40.65720720 3001.83457914 + ------------------------------------------------------------------------------------- + TOTAL 11.90711942 794.00135844 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4879044 +BPFP 0.3871 bits/point +EBPFP 0.3871 equivalent bits/point +MSE 794.001358 +---------------------- --------------------------------------------------------- +Time: 21.253s Load: 1.263s, Pack+Encode: 7.698s, Decode+Unpack: 12.292s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 794.0014 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00328-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst (34/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.273s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 340,220B, BPFP=0.2160 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 335,796B, BPFP=0.2131 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 780,188B, BPFP=0.4952 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 767,780B, BPFP=0.4873 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 879,476B, BPFP=0.5582 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 880,312B, BPFP=0.5588 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 556,804B, BPFP=0.3534 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 549,524B, BPFP=0.3488 +⌛️ [2/4] FRONTEND: Frontend time: 7.763s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.300s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14568599 4.68819949 + layer.9.1 0.03329684 4.72242330 + layer.19.0 0.11848472 40.49695574 + layer.19.1 0.11973745 17.37726606 + layer.29.0 0.10886538 128.18800780 + layer.29.1 0.10946879 152.99416030 + layer.39.0 14.08931437 2948.16834579 + layer.39.1 9.95616799 2964.31134222 + ------------------------------------------------------------------------------------- + TOTAL 3.08512769 782.61833759 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5090100 +BPFP 0.4039 bits/point +EBPFP 0.4039 equivalent bits/point +MSE 782.618338 +---------------------- --------------------------------------------------------- +Time: 21.335s Load: 1.273s, Pack+Encode: 7.763s, Decode+Unpack: 12.300s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 782.6183 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00332-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst (35/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.256s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 338,596B, BPFP=0.2149 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 322,904B, BPFP=0.2050 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 679,880B, BPFP=0.4316 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 699,124B, BPFP=0.4438 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 866,080B, BPFP=0.5497 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 877,644B, BPFP=0.5571 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 542,588B, BPFP=0.3444 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 542,500B, BPFP=0.3444 +⌛️ [2/4] FRONTEND: Frontend time: 7.686s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.268s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14493624 4.63362212 + layer.9.1 0.14482686 4.58129018 + layer.19.0 0.11946148 12.33893987 + layer.19.1 0.12828579 17.37560047 + layer.29.0 0.10467725 100.05108466 + layer.29.1 0.10613328 177.41109441 + layer.39.0 22.00188902 2947.77543061 + layer.39.1 19.26198661 2833.59896003 + ------------------------------------------------------------------------------------- + TOTAL 5.25152457 762.22075279 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4869316 +BPFP 0.3863 bits/point +EBPFP 0.3863 equivalent bits/point +MSE 762.220753 +---------------------- --------------------------------------------------------- +Time: 21.210s Load: 1.256s, Pack+Encode: 7.686s, Decode+Unpack: 12.268s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 762.2208 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00333-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst (36/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.263s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 312,240B, BPFP=0.1982 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 333,836B, BPFP=0.2119 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 723,936B, BPFP=0.4595 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 674,156B, BPFP=0.4279 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 855,768B, BPFP=0.5432 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 832,276B, BPFP=0.5283 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 533,328B, BPFP=0.3385 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 572,220B, BPFP=0.3632 +⌛️ [2/4] FRONTEND: Frontend time: 7.710s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.303s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14475438 4.56375363 + layer.9.1 0.14492096 4.59941229 + layer.19.0 0.11744098 12.32180035 + layer.19.1 0.11578254 8.40969479 + layer.29.0 0.11402616 154.53270231 + layer.29.1 0.11062706 116.71771409 + layer.39.0 28.92800668 2827.55476113 + layer.39.1 10.80449708 2883.24959376 + ------------------------------------------------------------------------------------- + TOTAL 5.06000698 751.49367904 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4837760 +BPFP 0.3838 bits/point +EBPFP 0.3838 equivalent bits/point +MSE 751.493679 +---------------------- --------------------------------------------------------- +Time: 21.275s Load: 1.263s, Pack+Encode: 7.710s, Decode+Unpack: 12.303s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 751.4937 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00334-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst (37/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.264s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 311,080B, BPFP=0.1975 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 310,824B, BPFP=0.1973 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 598,364B, BPFP=0.3798 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 654,720B, BPFP=0.4156 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 727,904B, BPFP=0.4620 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 763,312B, BPFP=0.4845 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 490,260B, BPFP=0.3112 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 556,632B, BPFP=0.3533 +⌛️ [2/4] FRONTEND: Frontend time: 7.711s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.271s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14546214 4.62670494 + layer.9.1 0.14553630 4.65396807 + layer.19.0 0.04765745 26.83475940 + layer.19.1 0.04191649 17.09443553 + layer.29.0 0.16505912 130.36633694 + layer.29.1 0.15755973 143.79148724 + layer.39.0 42.51041751 2703.14283393 + layer.39.1 31.38856333 2755.80370491 + ------------------------------------------------------------------------------------- + TOTAL 9.32527151 723.28927887 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4413096 +BPFP 0.3502 bits/point +EBPFP 0.3502 equivalent bits/point +MSE 723.289279 +---------------------- --------------------------------------------------------- +Time: 21.246s Load: 1.264s, Pack+Encode: 7.711s, Decode+Unpack: 12.271s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 723.2893 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01347-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst (38/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.271s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 310,224B, BPFP=0.1969 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 336,636B, BPFP=0.2137 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 747,288B, BPFP=0.4743 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 717,776B, BPFP=0.4556 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 784,460B, BPFP=0.4979 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 789,880B, BPFP=0.5014 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 544,108B, BPFP=0.3454 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 553,228B, BPFP=0.3512 +⌛️ [2/4] FRONTEND: Frontend time: 7.690s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.290s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03343166 4.66861429 + layer.9.1 0.03311388 4.66996313 + layer.19.0 0.03842411 12.40563556 + layer.19.1 0.03806642 7.58462293 + layer.29.0 4.26870163 89.30802121 + layer.29.1 4.26552788 67.02912740 + layer.39.0 33.95300821 2534.89974001 + layer.39.1 48.19954501 2705.48293793 + ------------------------------------------------------------------------------------- + TOTAL 11.35372735 678.25608281 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4783600 +BPFP 0.3795 bits/point +EBPFP 0.3795 equivalent bits/point +MSE 678.256083 +---------------------- --------------------------------------------------------- +Time: 21.251s Load: 1.271s, Pack+Encode: 7.690s, Decode+Unpack: 12.290s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 678.2561 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01390-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst (39/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.258s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 330,632B, BPFP=0.2099 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 322,360B, BPFP=0.2046 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 734,828B, BPFP=0.4664 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 742,464B, BPFP=0.4713 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 797,424B, BPFP=0.5062 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 797,424B, BPFP=0.5062 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 547,780B, BPFP=0.3477 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 555,700B, BPFP=0.3527 +⌛️ [2/4] FRONTEND: Frontend time: 7.733s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.278s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14549638 4.62823564 + layer.9.1 0.14520178 4.66087479 + layer.19.0 0.11487435 12.77749228 + layer.19.1 0.11481158 7.75026786 + layer.29.0 0.10827909 97.09650227 + layer.29.1 0.10618535 78.93850544 + layer.39.0 9.83978281 2713.43776406 + layer.39.1 9.67554703 2805.58628534 + ------------------------------------------------------------------------------------- + TOTAL 2.53127230 715.60949096 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4828612 +BPFP 0.3831 bits/point +EBPFP 0.3831 equivalent bits/point +MSE 715.609491 +---------------------- --------------------------------------------------------- +Time: 21.269s Load: 1.258s, Pack+Encode: 7.733s, Decode+Unpack: 12.278s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 715.6095 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01413-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst (40/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.266s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 302,464B, BPFP=0.1920 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 308,552B, BPFP=0.1959 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 704,028B, BPFP=0.4469 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 678,316B, BPFP=0.4306 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 824,884B, BPFP=0.5236 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 800,468B, BPFP=0.5081 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 522,708B, BPFP=0.3318 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 528,008B, BPFP=0.3352 +⌛️ [2/4] FRONTEND: Frontend time: 7.702s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.335s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00092318 4.69667689 + layer.9.1 0.00095285 4.67545213 + layer.19.0 0.08568402 7.66319543 + layer.19.1 0.08404610 12.54386756 + layer.29.0 0.12100375 127.49214942 + layer.29.1 0.12795564 103.26310123 + layer.39.0 12.85620633 2937.68183295 + layer.39.1 12.98640239 2820.12642184 + ------------------------------------------------------------------------------------- + TOTAL 3.28289678 752.26783718 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4669428 +BPFP 0.3705 bits/point +EBPFP 0.3705 equivalent bits/point +MSE 752.267837 +---------------------- --------------------------------------------------------- +Time: 21.303s Load: 1.266s, Pack+Encode: 7.702s, Decode+Unpack: 12.335s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 752.2678 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01421-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst (41/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.258s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 318,736B, BPFP=0.2023 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 310,864B, BPFP=0.1973 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 693,364B, BPFP=0.4401 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 686,052B, BPFP=0.4355 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 819,204B, BPFP=0.5200 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 806,472B, BPFP=0.5119 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 541,516B, BPFP=0.3437 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 551,200B, BPFP=0.3499 +⌛️ [2/4] FRONTEND: Frontend time: 7.690s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.273s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00113376 4.66437289 + layer.9.1 0.00100095 4.70696714 + layer.19.0 0.00983371 7.65761471 + layer.19.1 0.00806405 7.86476720 + layer.29.0 4.28365570 52.37577693 + layer.29.1 4.28597952 101.87180086 + layer.39.0 8.41906814 2677.29411765 + layer.39.1 8.59662605 2873.83295418 + ------------------------------------------------------------------------------------- + TOTAL 3.20067024 716.28354645 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4727408 +BPFP 0.3751 bits/point +EBPFP 0.3751 equivalent bits/point +MSE 716.283546 +---------------------- --------------------------------------------------------- +Time: 21.221s Load: 1.258s, Pack+Encode: 7.690s, Decode+Unpack: 12.273s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 716.2835 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01422-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst (42/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.274s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 330,836B, BPFP=0.2100 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 325,756B, BPFP=0.2068 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 755,984B, BPFP=0.4799 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 768,512B, BPFP=0.4878 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 842,096B, BPFP=0.5345 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 844,484B, BPFP=0.5360 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 530,064B, BPFP=0.3365 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 550,328B, BPFP=0.3493 +⌛️ [2/4] FRONTEND: Frontend time: 7.681s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.291s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14550294 4.68184787 + layer.9.1 0.14526658 4.64389015 + layer.19.0 0.11599200 7.79102895 + layer.19.1 0.11361485 7.83676712 + layer.29.0 4.26439454 71.97537679 + layer.29.1 4.25587461 63.94406585 + layer.39.0 8.37236706 2800.06662333 + layer.39.1 8.35116642 2549.64007150 + ------------------------------------------------------------------------------------- + TOTAL 3.22052237 688.82245895 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4948060 +BPFP 0.3926 bits/point +EBPFP 0.3926 equivalent bits/point +MSE 688.822459 +---------------------- --------------------------------------------------------- +Time: 21.245s Load: 1.274s, Pack+Encode: 7.681s, Decode+Unpack: 12.291s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 688.8225 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01441-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst (43/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.263s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 325,652B, BPFP=0.2067 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 330,152B, BPFP=0.2096 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 773,900B, BPFP=0.4912 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 758,932B, BPFP=0.4817 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 843,944B, BPFP=0.5357 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 838,608B, BPFP=0.5323 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 563,132B, BPFP=0.3574 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 552,664B, BPFP=0.3508 +⌛️ [2/4] FRONTEND: Frontend time: 7.719s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.360s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00080404 4.69220508 + layer.9.1 0.00082438 4.68770598 + layer.19.0 0.00843097 17.24508196 + layer.19.1 0.00674472 7.66264701 + layer.29.0 4.27713270 52.58979424 + layer.29.1 4.27133426 84.24759303 + layer.39.0 22.97048921 2744.89892753 + layer.39.1 18.06488920 2843.01949951 + ------------------------------------------------------------------------------------- + TOTAL 6.20008118 719.88043179 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4986984 +BPFP 0.3957 bits/point +EBPFP 0.3957 equivalent bits/point +MSE 719.880432 +---------------------- --------------------------------------------------------- +Time: 21.341s Load: 1.263s, Pack+Encode: 7.719s, Decode+Unpack: 12.360s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 719.8804 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00350-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst (44/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.261s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 326,928B, BPFP=0.2075 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 318,516B, BPFP=0.2022 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 723,584B, BPFP=0.4593 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 735,944B, BPFP=0.4671 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 839,768B, BPFP=0.5330 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 841,232B, BPFP=0.5340 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 573,752B, BPFP=0.3642 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 559,012B, BPFP=0.3548 +⌛️ [2/4] FRONTEND: Frontend time: 7.716s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.352s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14506060 4.65392745 + layer.9.1 0.14523201 4.64180468 + layer.19.0 0.04621643 7.48327180 + layer.19.1 0.04629335 11.85916553 + layer.29.0 4.27940669 83.80066319 + layer.29.1 4.27759670 73.73114539 + layer.39.0 19.91382637 2672.19954501 + layer.39.1 24.01088215 2742.22326942 + ------------------------------------------------------------------------------------- + TOTAL 6.60806429 700.07409906 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4918736 +BPFP 0.3903 bits/point +EBPFP 0.3903 equivalent bits/point +MSE 700.074099 +---------------------- --------------------------------------------------------- +Time: 21.329s Load: 1.261s, Pack+Encode: 7.716s, Decode+Unpack: 12.352s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 700.0741 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00351-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst (45/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.255s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 288,120B, BPFP=0.1829 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 291,344B, BPFP=0.1849 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 511,600B, BPFP=0.3247 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 519,732B, BPFP=0.3299 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 708,880B, BPFP=0.4500 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 683,644B, BPFP=0.4339 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 480,416B, BPFP=0.3049 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 497,628B, BPFP=0.3159 +⌛️ [2/4] FRONTEND: Frontend time: 7.696s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.338s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66985745 4.70922558 + layer.9.1 2.66884121 4.73700191 + layer.19.0 3.21935619 8.77912803 + layer.19.1 3.21606501 8.47198088 + layer.29.0 4.24164606 51.12364417 + layer.29.1 4.23648681 84.97366042 + layer.39.0 8.06392628 2462.82157946 + layer.39.1 8.17747540 2343.69499513 + ------------------------------------------------------------------------------------- + TOTAL 4.56170680 621.16390195 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 3981364 +BPFP 0.3159 bits/point +EBPFP 0.3159 equivalent bits/point +MSE 621.163902 +---------------------- --------------------------------------------------------- +Time: 21.289s Load: 1.255s, Pack+Encode: 7.696s, Decode+Unpack: 12.338s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 621.1639 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00358-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst (46/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.269s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 289,664B, BPFP=0.1839 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 288,156B, BPFP=0.1829 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 561,628B, BPFP=0.3565 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 571,424B, BPFP=0.3627 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 737,504B, BPFP=0.4681 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 766,912B, BPFP=0.4868 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 508,364B, BPFP=0.3227 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 508,364B, BPFP=0.3227 +⌛️ [2/4] FRONTEND: Frontend time: 7.778s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.258s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66795352 4.84302480 + layer.9.1 2.66862889 4.74268990 + layer.19.0 3.22250645 9.47153528 + layer.19.1 3.22577319 13.50213911 + layer.29.0 4.25792136 77.27293732 + layer.29.1 4.25014663 81.25555533 + layer.39.0 8.65209937 2603.84465388 + layer.39.1 8.58450170 2767.52778681 + ------------------------------------------------------------------------------------- + TOTAL 4.69119139 695.30754030 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4232016 +BPFP 0.3358 bits/point +EBPFP 0.3358 equivalent bits/point +MSE 695.307540 +---------------------- --------------------------------------------------------- +Time: 21.305s Load: 1.269s, Pack+Encode: 7.778s, Decode+Unpack: 12.258s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 695.3075 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00360-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst (47/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.262s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 298,884B, BPFP=0.1897 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 300,656B, BPFP=0.1908 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 756,836B, BPFP=0.4804 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 762,868B, BPFP=0.4842 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 849,656B, BPFP=0.5393 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 846,944B, BPFP=0.5376 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 527,404B, BPFP=0.3348 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 518,208B, BPFP=0.3289 +⌛️ [2/4] FRONTEND: Frontend time: 7.752s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.340s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00092775 4.69840468 + layer.9.1 0.00093166 4.71447433 + layer.19.0 0.08227225 16.55712240 + layer.19.1 0.08381199 12.27941430 + layer.29.0 0.10725604 72.07693167 + layer.29.1 0.10756977 58.22669199 + layer.39.0 7.96294394 2759.95190120 + layer.39.1 7.95922050 2838.13454664 + ------------------------------------------------------------------------------------- + TOTAL 2.03811674 720.82993590 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4861456 +BPFP 0.3857 bits/point +EBPFP 0.3857 equivalent bits/point +MSE 720.829936 +---------------------- --------------------------------------------------------- +Time: 21.354s Load: 1.262s, Pack+Encode: 7.752s, Decode+Unpack: 12.340s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 720.8299 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00361-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst (48/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.262s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 286,380B, BPFP=0.1818 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 291,420B, BPFP=0.1850 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 619,400B, BPFP=0.3932 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 588,164B, BPFP=0.3733 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 730,876B, BPFP=0.4639 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 715,896B, BPFP=0.4544 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 507,096B, BPFP=0.3219 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 508,976B, BPFP=0.3231 +⌛️ [2/4] FRONTEND: Frontend time: 7.736s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.307s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66308118 4.81437886 + layer.9.1 2.66351027 4.82242114 + layer.19.0 3.21594155 9.02658269 + layer.19.1 3.21498593 9.00418871 + layer.29.0 4.33566519 126.18903356 + layer.29.1 4.34101296 118.97539202 + layer.39.0 8.65310735 2575.90867728 + layer.39.1 8.66575030 2556.01592460 + ------------------------------------------------------------------------------------- + TOTAL 4.71913184 675.59457486 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4248208 +BPFP 0.3371 bits/point +EBPFP 0.3371 equivalent bits/point +MSE 675.594575 +---------------------- --------------------------------------------------------- +Time: 21.306s Load: 1.262s, Pack+Encode: 7.736s, Decode+Unpack: 12.307s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 675.5946 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00384-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst (49/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.257s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 296,864B, BPFP=0.1884 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 298,020B, BPFP=0.1892 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 628,660B, BPFP=0.3990 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 610,116B, BPFP=0.3873 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 699,688B, BPFP=0.4441 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 699,272B, BPFP=0.4439 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 507,640B, BPFP=0.3222 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 488,636B, BPFP=0.3102 +⌛️ [2/4] FRONTEND: Frontend time: 7.721s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.247s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.65959259 4.78085899 + layer.9.1 2.65993726 4.81470798 + layer.19.0 3.20866700 8.00749766 + layer.19.1 3.21007805 7.78942050 + layer.29.0 4.27255361 175.21347904 + layer.29.1 4.27602442 130.84345548 + layer.39.0 19.11658068 2186.68199545 + layer.39.1 9.60360322 2258.04322392 + ------------------------------------------------------------------------------------- + TOTAL 6.12587960 597.02182988 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4228896 +BPFP 0.3355 bits/point +EBPFP 0.3355 equivalent bits/point +MSE 597.021830 +---------------------- --------------------------------------------------------- +Time: 21.226s Load: 1.257s, Pack+Encode: 7.721s, Decode+Unpack: 12.247s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 597.0218 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00385-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst (50/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.262s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 307,736B, BPFP=0.1953 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 294,304B, BPFP=0.1868 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 681,096B, BPFP=0.4323 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 686,120B, BPFP=0.4355 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 737,976B, BPFP=0.4684 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 734,008B, BPFP=0.4659 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 519,488B, BPFP=0.3297 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 506,592B, BPFP=0.3216 +⌛️ [2/4] FRONTEND: Frontend time: 7.785s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.257s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67078767 4.68285236 + layer.9.1 2.67131261 4.70940712 + layer.19.0 3.30595795 7.75646747 + layer.19.1 3.30543206 7.90991631 + layer.29.0 0.11228124 93.40968273 + layer.29.1 0.11507649 123.88486147 + layer.39.0 11.41791162 2798.80760481 + layer.39.1 11.38150745 2863.37601560 + ------------------------------------------------------------------------------------- + TOTAL 4.37253339 738.06710098 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4467320 +BPFP 0.3545 bits/point +EBPFP 0.3545 equivalent bits/point +MSE 738.067101 +---------------------- --------------------------------------------------------- +Time: 21.305s Load: 1.262s, Pack+Encode: 7.785s, Decode+Unpack: 12.257s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 738.0671 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00557-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst (51/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.261s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 331,892B, BPFP=0.2107 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 322,460B, BPFP=0.2047 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 706,872B, BPFP=0.4487 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 652,636B, BPFP=0.4143 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 798,056B, BPFP=0.5066 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 795,068B, BPFP=0.5047 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 559,316B, BPFP=0.3550 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 540,312B, BPFP=0.3430 +⌛️ [2/4] FRONTEND: Frontend time: 7.733s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.256s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14463971 4.59922821 + layer.9.1 0.14470460 4.59887053 + layer.19.0 0.12255537 12.29327216 + layer.19.1 0.11825690 7.68363057 + layer.29.0 0.11949990 138.65240088 + layer.29.1 0.11467140 144.15625000 + layer.39.0 10.68243977 2641.31849204 + layer.39.1 10.40156301 2648.72798180 + ------------------------------------------------------------------------------------- + TOTAL 2.73104133 700.25376577 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4706612 +BPFP 0.3734 bits/point +EBPFP 0.3734 equivalent bits/point +MSE 700.253766 +---------------------- --------------------------------------------------------- +Time: 21.251s Load: 1.261s, Pack+Encode: 7.733s, Decode+Unpack: 12.256s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 700.2538 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00001-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst (52/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.261s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 322,860B, BPFP=0.2049 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 327,260B, BPFP=0.2077 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 735,904B, BPFP=0.4671 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 711,068B, BPFP=0.4514 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 846,404B, BPFP=0.5373 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 815,188B, BPFP=0.5174 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 574,164B, BPFP=0.3645 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 553,868B, BPFP=0.3516 +⌛️ [2/4] FRONTEND: Frontend time: 7.707s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.285s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14485273 4.55341197 + layer.9.1 0.14484227 4.55531495 + layer.19.0 0.11969613 21.61988646 + layer.19.1 0.11916645 27.66132038 + layer.29.0 0.11480527 68.76359888 + layer.29.1 0.11451660 115.42118947 + layer.39.0 11.00270276 2919.33734157 + layer.39.1 11.01557422 2867.96555086 + ------------------------------------------------------------------------------------- + TOTAL 2.84701956 753.73470182 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4886716 +BPFP 0.3877 bits/point +EBPFP 0.3877 equivalent bits/point +MSE 753.734702 +---------------------- --------------------------------------------------------- +Time: 21.253s Load: 1.261s, Pack+Encode: 7.707s, Decode+Unpack: 12.285s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 753.7347 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00126-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst (53/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.256s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 313,812B, BPFP=0.1992 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 307,836B, BPFP=0.1954 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 612,216B, BPFP=0.3886 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 594,048B, BPFP=0.3771 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 738,108B, BPFP=0.4685 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 738,804B, BPFP=0.4690 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 552,732B, BPFP=0.3508 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 507,276B, BPFP=0.3220 +⌛️ [2/4] FRONTEND: Frontend time: 7.715s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.271s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14481488 4.66169076 + layer.9.1 0.14470567 4.65641980 + layer.19.0 0.03819180 8.51081931 + layer.19.1 0.04002141 13.70486041 + layer.29.0 0.11241068 145.68101032 + layer.29.1 0.11133552 116.58685408 + layer.39.0 31.78807483 3030.82320442 + layer.39.1 43.50691623 3034.62398440 + ------------------------------------------------------------------------------------- + TOTAL 9.48580888 794.90610544 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4364832 +BPFP 0.3463 bits/point +EBPFP 0.3463 equivalent bits/point +MSE 794.906105 +---------------------- --------------------------------------------------------- +Time: 21.242s Load: 1.256s, Pack+Encode: 7.715s, Decode+Unpack: 12.271s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 794.9061 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00127-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst (54/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.298s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 320,380B, BPFP=0.2034 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 311,068B, BPFP=0.1975 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 667,156B, BPFP=0.4235 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 644,692B, BPFP=0.4092 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 810,740B, BPFP=0.5146 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 800,696B, BPFP=0.5082 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 543,688B, BPFP=0.3451 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 555,596B, BPFP=0.3527 +⌛️ [2/4] FRONTEND: Frontend time: 7.748s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.321s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14528272 4.60051262 + layer.9.1 0.14516892 4.61706372 + layer.19.0 0.11319376 8.13121991 + layer.19.1 0.11666145 30.82294798 + layer.29.0 0.21118872 272.77532905 + layer.29.1 0.20646930 193.87175008 + layer.39.0 14.37750853 3222.95937602 + layer.39.1 21.76644002 3222.86317842 + ------------------------------------------------------------------------------------- + TOTAL 4.63523918 870.08017223 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4654016 +BPFP 0.3693 bits/point +EBPFP 0.3693 equivalent bits/point +MSE 870.080172 +---------------------- --------------------------------------------------------- +Time: 21.368s Load: 1.298s, Pack+Encode: 7.748s, Decode+Unpack: 12.321s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 870.0802 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00128-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst (55/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.259s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 315,664B, BPFP=0.2004 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 314,024B, BPFP=0.1993 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 701,128B, BPFP=0.4450 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 727,204B, BPFP=0.4616 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 815,316B, BPFP=0.5175 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 804,888B, BPFP=0.5109 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 536,404B, BPFP=0.3405 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 539,936B, BPFP=0.3427 +⌛️ [2/4] FRONTEND: Frontend time: 7.682s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.305s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14454765 4.67574475 + layer.9.1 0.14475082 4.68284125 + layer.19.0 0.04087094 31.97577287 + layer.19.1 0.11687931 21.57040898 + layer.29.0 0.10817139 111.41628412 + layer.29.1 0.10802081 116.56664365 + layer.39.0 19.80422286 3108.53006175 + layer.39.1 34.29222355 2930.16834579 + ------------------------------------------------------------------------------------- + TOTAL 6.84496092 791.19826290 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4754564 +BPFP 0.3772 bits/point +EBPFP 0.3772 equivalent bits/point +MSE 791.198263 +---------------------- --------------------------------------------------------- +Time: 21.246s Load: 1.259s, Pack+Encode: 7.682s, Decode+Unpack: 12.305s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 791.1983 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00131-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst (56/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.256s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 316,180B, BPFP=0.2007 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 309,020B, BPFP=0.1962 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 587,508B, BPFP=0.3729 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 644,752B, BPFP=0.4093 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 725,288B, BPFP=0.4604 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 760,936B, BPFP=0.4830 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 509,072B, BPFP=0.3231 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 520,936B, BPFP=0.3307 +⌛️ [2/4] FRONTEND: Frontend time: 7.708s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.308s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14462205 4.72792404 + layer.9.1 0.14495783 4.74321103 + layer.19.0 0.04322015 49.65187784 + layer.19.1 0.03788725 9.15307752 + layer.29.0 0.10021623 82.54828161 + layer.29.1 0.10137775 66.45598391 + layer.39.0 58.66958482 2599.39389015 + layer.39.1 72.48303949 2736.60773481 + ------------------------------------------------------------------------------------- + TOTAL 16.46561320 694.16024761 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4373692 +BPFP 0.3470 bits/point +EBPFP 0.3470 equivalent bits/point +MSE 694.160248 +---------------------- --------------------------------------------------------- +Time: 21.272s Load: 1.256s, Pack+Encode: 7.708s, Decode+Unpack: 12.308s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 694.1602 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00132-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst (57/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.259s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 320,060B, BPFP=0.2032 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 318,992B, BPFP=0.2025 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 704,940B, BPFP=0.4475 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 671,128B, BPFP=0.4260 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 802,436B, BPFP=0.5093 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 813,464B, BPFP=0.5163 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 530,484B, BPFP=0.3367 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 543,368B, BPFP=0.3449 +⌛️ [2/4] FRONTEND: Frontend time: 7.695s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.254s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14527927 4.57903840 + layer.9.1 0.14528875 4.58582102 + layer.19.0 0.12591341 44.72856069 + layer.19.1 0.13556211 12.29943304 + layer.29.0 0.11238900 179.93571254 + layer.29.1 0.11028371 145.58879387 + layer.39.0 11.48751193 2642.59928502 + layer.39.1 11.29491489 2598.33376666 + ------------------------------------------------------------------------------------- + TOTAL 2.94464289 704.08130141 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4704872 +BPFP 0.3733 bits/point +EBPFP 0.3733 equivalent bits/point +MSE 704.081301 +---------------------- --------------------------------------------------------- +Time: 21.208s Load: 1.259s, Pack+Encode: 7.695s, Decode+Unpack: 12.254s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 704.0813 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00136-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst (58/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.266s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 323,936B, BPFP=0.2056 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 327,160B, BPFP=0.2077 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 709,928B, BPFP=0.4506 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 702,472B, BPFP=0.4459 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 778,884B, BPFP=0.4944 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 776,596B, BPFP=0.4929 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 540,376B, BPFP=0.3430 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 545,736B, BPFP=0.3464 +⌛️ [2/4] FRONTEND: Frontend time: 7.785s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.272s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14489184 4.60770243 + layer.9.1 0.14511764 4.61854586 + layer.19.0 0.03976490 21.85464231 + layer.19.1 0.11370806 12.27351367 + layer.29.0 0.10933599 54.45951312 + layer.29.1 0.11012027 95.35612000 + layer.39.0 9.10787636 2528.38804030 + layer.39.1 9.00026152 2462.89064023 + ------------------------------------------------------------------------------------- + TOTAL 2.34638457 648.05608974 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4705088 +BPFP 0.3733 bits/point +EBPFP 0.3733 equivalent bits/point +MSE 648.056090 +---------------------- --------------------------------------------------------- +Time: 21.322s Load: 1.266s, Pack+Encode: 7.785s, Decode+Unpack: 12.272s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 648.0561 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00193-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst (59/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.260s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 294,592B, BPFP=0.1870 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 299,388B, BPFP=0.1900 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 648,980B, BPFP=0.4119 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 616,232B, BPFP=0.3912 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 680,156B, BPFP=0.4317 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 671,248B, BPFP=0.4261 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 475,836B, BPFP=0.3020 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 468,144B, BPFP=0.2972 +⌛️ [2/4] FRONTEND: Frontend time: 7.747s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.289s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00083307 4.64499811 + layer.9.1 0.00247171 4.63432638 + layer.19.0 0.00642632 17.34983979 + layer.19.1 0.00641681 8.02788963 + layer.29.0 0.10256791 54.08377681 + layer.29.1 0.10162673 44.99216465 + layer.39.0 8.50517638 2088.72148196 + layer.39.1 8.55767781 2178.99090023 + ------------------------------------------------------------------------------------- + TOTAL 2.16039959 550.18067220 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4154576 +BPFP 0.3296 bits/point +EBPFP 0.3296 equivalent bits/point +MSE 550.180672 +---------------------- --------------------------------------------------------- +Time: 21.295s Load: 1.260s, Pack+Encode: 7.747s, Decode+Unpack: 12.289s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 550.1807 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00196-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst (60/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.261s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 309,596B, BPFP=0.1965 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 318,828B, BPFP=0.2024 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 647,064B, BPFP=0.4107 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 618,648B, BPFP=0.3927 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 788,324B, BPFP=0.5004 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 789,900B, BPFP=0.5014 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 532,008B, BPFP=0.3377 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 537,820B, BPFP=0.3414 +⌛️ [2/4] FRONTEND: Frontend time: 7.755s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.266s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00064391 4.68447669 + layer.9.1 0.00065402 4.66079417 + layer.19.0 0.08134466 8.12497905 + layer.19.1 0.08141702 11.99194883 + layer.29.0 0.11551180 166.84993500 + layer.29.1 0.11251285 107.36308702 + layer.39.0 10.61319619 2717.06386090 + layer.39.1 10.43102047 2744.88365291 + ------------------------------------------------------------------------------------- + TOTAL 2.67953762 720.70284182 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4542188 +BPFP 0.3604 bits/point +EBPFP 0.3604 equivalent bits/point +MSE 720.702842 +---------------------- --------------------------------------------------------- +Time: 21.282s Load: 1.261s, Pack+Encode: 7.755s, Decode+Unpack: 12.266s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 720.7028 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00559-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst (61/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.262s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 340,732B, BPFP=0.2163 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 331,880B, BPFP=0.2107 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 689,040B, BPFP=0.4374 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 720,556B, BPFP=0.4574 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 835,248B, BPFP=0.5302 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 834,660B, BPFP=0.5298 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 564,880B, BPFP=0.3586 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 571,652B, BPFP=0.3629 +⌛️ [2/4] FRONTEND: Frontend time: 7.738s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.289s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14441776 4.63976523 + layer.9.1 0.14449203 4.60670239 + layer.19.0 0.11315974 26.76814623 + layer.19.1 0.11435745 25.91604800 + layer.29.0 0.12811458 169.95910180 + layer.29.1 0.12952277 184.95316055 + layer.39.0 31.10682331 2925.64575886 + layer.39.1 16.99297713 3157.68053299 + ------------------------------------------------------------------------------------- + TOTAL 6.10923309 812.52115200 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4888648 +BPFP 0.3879 bits/point +EBPFP 0.3879 equivalent bits/point +MSE 812.521152 +---------------------- --------------------------------------------------------- +Time: 21.290s Load: 1.262s, Pack+Encode: 7.738s, Decode+Unpack: 12.289s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 812.5212 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00570-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst (62/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.269s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 309,564B, BPFP=0.1965 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 314,200B, BPFP=0.1994 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 659,776B, BPFP=0.4188 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 658,748B, BPFP=0.4181 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 755,796B, BPFP=0.4797 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 767,716B, BPFP=0.4873 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 526,560B, BPFP=0.3342 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 528,764B, BPFP=0.3356 +⌛️ [2/4] FRONTEND: Frontend time: 7.716s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.275s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00078516 4.73117808 + layer.9.1 0.00079184 4.73527381 + layer.19.0 3.22632161 7.57154201 + layer.19.1 3.22513146 12.49529397 + layer.29.0 0.10494786 125.56786237 + layer.29.1 0.10251782 58.68684494 + layer.39.0 10.88842496 2543.34952876 + layer.39.1 10.78217420 2679.25918102 + ------------------------------------------------------------------------------------- + TOTAL 3.54138686 679.54958812 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4521124 +BPFP 0.3587 bits/point +EBPFP 0.3587 equivalent bits/point +MSE 679.549588 +---------------------- --------------------------------------------------------- +Time: 21.261s Load: 1.269s, Pack+Encode: 7.716s, Decode+Unpack: 12.275s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 679.5496 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00758-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst (63/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.263s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 307,256B, BPFP=0.1950 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 325,312B, BPFP=0.2065 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 607,540B, BPFP=0.3856 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 590,076B, BPFP=0.3746 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 744,192B, BPFP=0.4724 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 740,392B, BPFP=0.4700 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 528,464B, BPFP=0.3354 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 520,708B, BPFP=0.3305 +⌛️ [2/4] FRONTEND: Frontend time: 7.753s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.356s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14561636 4.69685177 + layer.9.1 0.14552785 4.69577238 + layer.19.0 0.04069186 7.84643372 + layer.19.1 0.03840616 8.51957187 + layer.29.0 0.11346353 88.03281402 + layer.29.1 0.11182956 87.48982369 + layer.39.0 10.19697364 2320.26421839 + layer.39.1 10.11578978 2205.63422164 + ------------------------------------------------------------------------------------- + TOTAL 2.61353734 590.89746344 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4363940 +BPFP 0.3463 bits/point +EBPFP 0.3463 equivalent bits/point +MSE 590.897463 +---------------------- --------------------------------------------------------- +Time: 21.372s Load: 1.263s, Pack+Encode: 7.753s, Decode+Unpack: 12.356s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 590.8975 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00764-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst (64/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.259s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 308,244B, BPFP=0.1957 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 321,796B, BPFP=0.2043 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 711,192B, BPFP=0.4514 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 671,336B, BPFP=0.4261 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 814,808B, BPFP=0.5172 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 792,580B, BPFP=0.5031 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 546,232B, BPFP=0.3467 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 538,476B, BPFP=0.3418 +⌛️ [2/4] FRONTEND: Frontend time: 7.746s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.295s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14399445 4.70655106 + layer.9.1 0.14558028 4.65919175 + layer.19.0 0.03837104 31.42946661 + layer.19.1 0.04376782 49.74896917 + layer.29.0 0.11695251 135.31946701 + layer.29.1 0.13128335 124.82754103 + layer.39.0 11.28613757 2636.25138122 + layer.39.1 11.84408769 2715.01787455 + ------------------------------------------------------------------------------------- + TOTAL 2.96877184 712.74505530 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4704664 +BPFP 0.3733 bits/point +EBPFP 0.3733 equivalent bits/point +MSE 712.745055 +---------------------- --------------------------------------------------------- +Time: 21.301s Load: 1.259s, Pack+Encode: 7.746s, Decode+Unpack: 12.295s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 712.7451 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00845-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst (65/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.260s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 313,348B, BPFP=0.1989 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 330,500B, BPFP=0.2098 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 792,488B, BPFP=0.5030 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 715,412B, BPFP=0.4541 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 913,248B, BPFP=0.5797 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 907,072B, BPFP=0.5758 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 578,228B, BPFP=0.3670 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 603,492B, BPFP=0.3831 +⌛️ [2/4] FRONTEND: Frontend time: 7.794s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.279s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14502686 4.63157950 + layer.9.1 0.03259508 4.66587693 + layer.19.0 0.11326540 7.66160221 + layer.19.1 0.11324834 16.42967354 + layer.29.0 0.12250664 92.42518078 + layer.29.1 0.12058897 161.04131459 + layer.39.0 16.17915050 2991.09002275 + layer.39.1 21.66230805 3059.10139747 + ------------------------------------------------------------------------------------- + TOTAL 4.81108623 792.13083097 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5153788 +BPFP 0.4089 bits/point +EBPFP 0.4089 equivalent bits/point +MSE 792.130831 +---------------------- --------------------------------------------------------- +Time: 21.333s Load: 1.260s, Pack+Encode: 7.794s, Decode+Unpack: 12.279s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 792.1308 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00870-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst (66/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.266s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 300,216B, BPFP=0.1906 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 293,792B, BPFP=0.1865 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 618,484B, BPFP=0.3926 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 613,668B, BPFP=0.3895 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 815,836B, BPFP=0.5179 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 722,568B, BPFP=0.4586 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 517,108B, BPFP=0.3282 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 493,456B, BPFP=0.3132 +⌛️ [2/4] FRONTEND: Frontend time: 7.735s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.323s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66733019 4.73432454 + layer.9.1 2.66763138 4.73873319 + layer.19.0 3.22293078 8.14548718 + layer.19.1 3.22376992 8.52214325 + layer.29.0 4.27658332 59.92757759 + layer.29.1 4.27160529 55.99336306 + layer.39.0 7.81683598 2734.33734157 + layer.39.1 9.86231960 2817.64933377 + ------------------------------------------------------------------------------------- + TOTAL 4.75112581 711.75603802 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4375128 +BPFP 0.3471 bits/point +EBPFP 0.3471 equivalent bits/point +MSE 711.756038 +---------------------- --------------------------------------------------------- +Time: 21.325s Load: 1.266s, Pack+Encode: 7.735s, Decode+Unpack: 12.323s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 711.7560 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00153-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst (67/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.272s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 293,108B, BPFP=0.1861 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 292,756B, BPFP=0.1858 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 562,072B, BPFP=0.3568 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 590,532B, BPFP=0.3748 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 701,484B, BPFP=0.4453 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 752,252B, BPFP=0.4775 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 478,512B, BPFP=0.3037 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 497,576B, BPFP=0.3158 +⌛️ [2/4] FRONTEND: Frontend time: 7.721s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.261s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14549763 4.66256100 + layer.9.1 0.14520254 4.65705296 + layer.19.0 0.04746155 22.96322006 + layer.19.1 0.04383140 8.54686484 + layer.29.0 4.26247378 69.00806894 + layer.29.1 4.25497898 73.50802324 + layer.39.0 7.94138086 2432.90120247 + layer.39.1 7.86439079 2705.24601885 + ------------------------------------------------------------------------------------- + TOTAL 3.08815219 665.18662654 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4168292 +BPFP 0.3307 bits/point +EBPFP 0.3307 equivalent bits/point +MSE 665.186627 +---------------------- --------------------------------------------------------- +Time: 21.255s Load: 1.272s, Pack+Encode: 7.721s, Decode+Unpack: 12.261s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 665.1866 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00167-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst (68/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.262s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 300,612B, BPFP=0.1908 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 286,716B, BPFP=0.1820 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 523,536B, BPFP=0.3323 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 533,132B, BPFP=0.3384 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 666,628B, BPFP=0.4231 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 674,324B, BPFP=0.4280 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 498,696B, BPFP=0.3165 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 495,300B, BPFP=0.3144 +⌛️ [2/4] FRONTEND: Frontend time: 7.687s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.247s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11283708 4.70976893 + layer.9.1 0.11300174 4.70028196 + layer.19.0 3.22718329 9.10336075 + layer.19.1 3.22892155 8.80941067 + layer.29.0 4.26448309 81.80119638 + layer.29.1 4.25758082 48.30728490 + layer.39.0 9.82393946 2479.98277543 + layer.39.1 9.78394007 2558.17110822 + ------------------------------------------------------------------------------------- + TOTAL 4.35148589 649.44814840 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 3978944 +BPFP 0.3157 bits/point +EBPFP 0.3157 equivalent bits/point +MSE 649.448148 +---------------------- --------------------------------------------------------- +Time: 21.196s Load: 1.262s, Pack+Encode: 7.687s, Decode+Unpack: 12.247s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 649.4481 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00208-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst (69/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.258s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 323,944B, BPFP=0.2056 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 323,728B, BPFP=0.2055 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 681,168B, BPFP=0.4324 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 680,448B, BPFP=0.4319 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 794,696B, BPFP=0.5044 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 792,436B, BPFP=0.5030 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 515,996B, BPFP=0.3275 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 533,052B, BPFP=0.3384 +⌛️ [2/4] FRONTEND: Frontend time: 7.711s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.345s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14503861 4.54115210 + layer.9.1 0.14483112 4.56641482 + layer.19.0 0.11529889 8.02816385 + layer.19.1 0.11517203 31.29046403 + layer.29.0 0.11961639 94.02419158 + layer.29.1 0.11795276 92.45660343 + layer.39.0 83.84633978 2942.74130647 + layer.39.1 174.87768118 3055.63470913 + ------------------------------------------------------------------------------------- + TOTAL 32.43524135 779.16037568 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4645468 +BPFP 0.3686 bits/point +EBPFP 0.3686 equivalent bits/point +MSE 779.160376 +---------------------- --------------------------------------------------------- +Time: 21.314s Load: 1.258s, Pack+Encode: 7.711s, Decode+Unpack: 12.345s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 779.1604 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01201-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst (70/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.265s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 315,284B, BPFP=0.2001 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 315,708B, BPFP=0.2004 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 631,616B, BPFP=0.4009 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 660,392B, BPFP=0.4192 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 819,276B, BPFP=0.5200 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 775,236B, BPFP=0.4921 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 558,300B, BPFP=0.3544 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 586,748B, BPFP=0.3724 +⌛️ [2/4] FRONTEND: Frontend time: 7.727s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.289s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.81453913 4.70981748 + layer.9.1 0.14528001 4.64508412 + layer.19.0 3.26598681 22.06569660 + layer.19.1 0.04116655 21.88536419 + layer.29.0 4.28557138 96.15530549 + layer.29.1 4.28198282 110.04127397 + layer.39.0 74.89367180 2615.46815080 + layer.39.1 42.04871577 2616.85570361 + ------------------------------------------------------------------------------------- + TOTAL 16.47211428 686.47829953 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4662560 +BPFP 0.3699 bits/point +EBPFP 0.3699 equivalent bits/point +MSE 686.478300 +---------------------- --------------------------------------------------------- +Time: 21.281s Load: 1.265s, Pack+Encode: 7.727s, Decode+Unpack: 12.289s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 686.4783 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01296-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst (71/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.269s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 316,020B, BPFP=0.2006 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 314,940B, BPFP=0.1999 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 664,132B, BPFP=0.4216 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 676,904B, BPFP=0.4297 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 738,960B, BPFP=0.4691 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 743,968B, BPFP=0.4722 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 505,412B, BPFP=0.3208 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 521,432B, BPFP=0.3310 +⌛️ [2/4] FRONTEND: Frontend time: 7.723s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.275s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66710422 4.81660080 + layer.9.1 2.66812426 4.73358125 + layer.19.0 3.22059776 36.31433062 + layer.19.1 3.22546153 8.00727741 + layer.29.0 0.11226317 93.86950561 + layer.29.1 0.11257672 131.66579664 + layer.39.0 59.39237691 2558.22814430 + layer.39.1 37.52358222 2565.15680858 + ------------------------------------------------------------------------------------- + TOTAL 13.61526085 675.34900565 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4481768 +BPFP 0.3556 bits/point +EBPFP 0.3556 equivalent bits/point +MSE 675.349006 +---------------------- --------------------------------------------------------- +Time: 21.266s Load: 1.269s, Pack+Encode: 7.723s, Decode+Unpack: 12.275s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 675.3490 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01313-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst (72/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.258s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 306,236B, BPFP=0.1944 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 305,364B, BPFP=0.1938 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 691,276B, BPFP=0.4388 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 639,088B, BPFP=0.4057 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 772,000B, BPFP=0.4900 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 813,244B, BPFP=0.5162 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 499,916B, BPFP=0.3173 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 500,368B, BPFP=0.3176 +⌛️ [2/4] FRONTEND: Frontend time: 7.725s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.250s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14519123 4.66478516 + layer.9.1 0.14511500 4.63806664 + layer.19.0 0.03974548 8.17126818 + layer.19.1 0.03981401 13.57979058 + layer.29.0 4.26343511 79.97101479 + layer.29.1 4.25610090 103.68936870 + layer.39.0 7.90972018 2447.55216120 + layer.39.1 8.05601540 2411.30581735 + ------------------------------------------------------------------------------------- + TOTAL 3.10689216 634.19653408 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4527492 +BPFP 0.3592 bits/point +EBPFP 0.3592 equivalent bits/point +MSE 634.196534 +---------------------- --------------------------------------------------------- +Time: 21.234s Load: 1.258s, Pack+Encode: 7.725s, Decode+Unpack: 12.250s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 634.1965 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00032-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst (73/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.298s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 313,272B, BPFP=0.1988 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 299,752B, BPFP=0.1903 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 615,472B, BPFP=0.3907 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 619,040B, BPFP=0.3929 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 772,460B, BPFP=0.4903 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 759,244B, BPFP=0.4819 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 488,412B, BPFP=0.3100 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 488,248B, BPFP=0.3099 +⌛️ [2/4] FRONTEND: Frontend time: 7.719s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.243s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14586543 4.70402571 + layer.9.1 0.14572574 4.68770185 + layer.19.0 0.03953905 17.60432239 + layer.19.1 0.03760033 8.14926458 + layer.29.0 0.10448607 65.95397303 + layer.29.1 0.10697372 54.72960676 + layer.39.0 14.19073468 2320.57637309 + layer.39.1 8.92149669 2428.08011050 + ------------------------------------------------------------------------------------- + TOTAL 2.96155271 613.06067224 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4355900 +BPFP 0.3456 bits/point +EBPFP 0.3456 equivalent bits/point +MSE 613.060672 +---------------------- --------------------------------------------------------- +Time: 21.260s Load: 1.298s, Pack+Encode: 7.719s, Decode+Unpack: 12.243s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 613.0607 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00041-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst (74/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.268s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 318,960B, BPFP=0.2025 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 319,264B, BPFP=0.2027 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 737,124B, BPFP=0.4679 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 712,824B, BPFP=0.4525 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 780,964B, BPFP=0.4957 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 790,168B, BPFP=0.5016 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 514,868B, BPFP=0.3268 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 523,868B, BPFP=0.3325 +⌛️ [2/4] FRONTEND: Frontend time: 7.733s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.291s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00251883 4.71218193 + layer.9.1 0.14409062 4.69235964 + layer.19.0 0.12740102 68.09901081 + layer.19.1 0.12254588 26.42263162 + layer.29.0 4.25147928 91.99491185 + layer.29.1 4.25065697 59.92692761 + layer.39.0 9.21805114 2686.96928827 + layer.39.1 9.03214690 2751.50796230 + ------------------------------------------------------------------------------------- + TOTAL 3.39361133 711.79065925 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4698040 +BPFP 0.3728 bits/point +EBPFP 0.3728 equivalent bits/point +MSE 711.790659 +---------------------- --------------------------------------------------------- +Time: 21.292s Load: 1.268s, Pack+Encode: 7.733s, Decode+Unpack: 12.291s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 711.7907 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00042-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst (75/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.270s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 340,992B, BPFP=0.2164 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 332,644B, BPFP=0.2111 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 765,032B, BPFP=0.4856 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 800,848B, BPFP=0.5083 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 876,036B, BPFP=0.5561 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 887,272B, BPFP=0.5632 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 563,816B, BPFP=0.3579 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 590,716B, BPFP=0.3750 +⌛️ [2/4] FRONTEND: Frontend time: 7.721s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.349s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14550905 4.70252104 + layer.9.1 0.14590163 4.68692460 + layer.19.0 0.12839093 31.70478804 + layer.19.1 0.12422524 12.77364951 + layer.29.0 0.11695262 165.48898074 + layer.29.1 0.11389293 151.60993866 + layer.39.0 10.18180439 2950.52388690 + layer.39.1 10.42432323 2858.12674683 + ------------------------------------------------------------------------------------- + TOTAL 2.67262500 772.45217954 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5157356 +BPFP 0.4092 bits/point +EBPFP 0.4092 equivalent bits/point +MSE 772.452180 +---------------------- --------------------------------------------------------- +Time: 21.340s Load: 1.270s, Pack+Encode: 7.721s, Decode+Unpack: 12.349s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 772.4522 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00620-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst (76/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.256s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 309,716B, BPFP=0.1966 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 339,392B, BPFP=0.2154 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 743,036B, BPFP=0.4716 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 756,340B, BPFP=0.4801 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 829,236B, BPFP=0.5264 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 827,312B, BPFP=0.5251 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 531,588B, BPFP=0.3374 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 550,428B, BPFP=0.3494 +⌛️ [2/4] FRONTEND: Frontend time: 7.728s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.283s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14509775 4.70220080 + layer.9.1 0.14508723 4.69445717 + layer.19.0 0.11633494 22.00146246 + layer.19.1 0.11804005 26.31693309 + layer.29.0 0.15409572 145.99446498 + layer.29.1 0.14997486 157.30459864 + layer.39.0 9.23291952 2827.61455964 + layer.39.1 9.22304726 2709.03509912 + ------------------------------------------------------------------------------------- + TOTAL 2.41057467 737.20797199 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4887048 +BPFP 0.3878 bits/point +EBPFP 0.3878 equivalent bits/point +MSE 737.207972 +---------------------- --------------------------------------------------------- +Time: 21.267s Load: 1.256s, Pack+Encode: 7.728s, Decode+Unpack: 12.283s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 737.2080 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00636-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst (77/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.264s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 328,756B, BPFP=0.2087 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 325,680B, BPFP=0.2067 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 750,636B, BPFP=0.4765 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 769,724B, BPFP=0.4886 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 846,176B, BPFP=0.5371 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 847,268B, BPFP=0.5378 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 565,424B, BPFP=0.3589 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 561,088B, BPFP=0.3562 +⌛️ [2/4] FRONTEND: Frontend time: 7.746s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.291s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14497060 4.64500128 + layer.9.1 0.14492971 4.62412373 + layer.19.0 0.11929473 7.63278076 + layer.19.1 0.11869117 35.16958482 + layer.29.0 0.13715227 89.21921718 + layer.29.1 0.14278979 133.21712504 + layer.39.0 9.99110525 2763.83327917 + layer.39.1 10.01170034 2587.09587260 + ------------------------------------------------------------------------------------- + TOTAL 2.60132923 703.17962307 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4994752 +BPFP 0.3963 bits/point +EBPFP 0.3963 equivalent bits/point +MSE 703.179623 +---------------------- --------------------------------------------------------- +Time: 21.302s Load: 1.264s, Pack+Encode: 7.746s, Decode+Unpack: 12.291s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 703.1796 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00637-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst (78/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.263s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 333,616B, BPFP=0.2118 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 324,824B, BPFP=0.2062 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 726,268B, BPFP=0.4610 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 690,172B, BPFP=0.4381 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 875,460B, BPFP=0.5557 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 839,048B, BPFP=0.5326 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 553,280B, BPFP=0.3512 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 589,172B, BPFP=0.3740 +⌛️ [2/4] FRONTEND: Frontend time: 7.726s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.302s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14589856 4.72290539 + layer.9.1 0.03321603 4.72370613 + layer.19.0 0.11866178 44.36664669 + layer.19.1 0.11267978 26.42069183 + layer.29.0 0.10803594 77.46911562 + layer.29.1 0.10714094 103.43180249 + layer.39.0 11.58943751 3038.33084173 + layer.39.1 9.70079103 3126.19759506 + ------------------------------------------------------------------------------------- + TOTAL 2.73948269 803.20791312 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4931840 +BPFP 0.3913 bits/point +EBPFP 0.3913 equivalent bits/point +MSE 803.207913 +---------------------- --------------------------------------------------------- +Time: 21.291s Load: 1.263s, Pack+Encode: 7.726s, Decode+Unpack: 12.302s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 803.2079 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00410-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst (79/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.265s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 317,984B, BPFP=0.2018 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 322,708B, BPFP=0.2048 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 698,180B, BPFP=0.4432 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 662,368B, BPFP=0.4204 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 786,096B, BPFP=0.4990 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 789,768B, BPFP=0.5013 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 541,192B, BPFP=0.3435 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 546,280B, BPFP=0.3468 +⌛️ [2/4] FRONTEND: Frontend time: 7.704s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.262s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03308301 4.72674499 + layer.9.1 0.14566304 4.68372451 + layer.19.0 0.03810260 12.60205380 + layer.19.1 0.03780774 7.54179192 + layer.29.0 0.11592613 74.28267184 + layer.29.1 0.11717217 85.99268768 + layer.39.0 9.98032847 2864.65745856 + layer.39.1 9.70849498 2691.72261943 + ------------------------------------------------------------------------------------- + TOTAL 2.52207227 718.27621909 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4664576 +BPFP 0.3701 bits/point +EBPFP 0.3701 equivalent bits/point +MSE 718.276219 +---------------------- --------------------------------------------------------- +Time: 21.231s Load: 1.265s, Pack+Encode: 7.704s, Decode+Unpack: 12.262s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 718.2762 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00411-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst (80/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.257s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 303,856B, BPFP=0.1929 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 305,140B, BPFP=0.1937 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 616,036B, BPFP=0.3910 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 561,100B, BPFP=0.3562 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 703,508B, BPFP=0.4466 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 706,532B, BPFP=0.4485 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 499,288B, BPFP=0.3169 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 497,616B, BPFP=0.3159 +⌛️ [2/4] FRONTEND: Frontend time: 7.703s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.283s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14555504 4.67021005 + layer.9.1 0.14557384 4.68147178 + layer.19.0 0.03995539 8.03525654 + layer.19.1 0.04542811 12.02390848 + layer.29.0 0.12033866 82.53742485 + layer.29.1 0.13252172 63.84507028 + layer.39.0 10.37566776 2258.96165096 + layer.39.1 9.84188447 2285.10578486 + ------------------------------------------------------------------------------------- + TOTAL 2.60586562 589.98259722 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4193076 +BPFP 0.3327 bits/point +EBPFP 0.3327 equivalent bits/point +MSE 589.982597 +---------------------- --------------------------------------------------------- +Time: 21.244s Load: 1.257s, Pack+Encode: 7.703s, Decode+Unpack: 12.283s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 589.9826 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00412-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst (81/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.266s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 334,432B, BPFP=0.2123 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 339,788B, BPFP=0.2157 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 750,312B, BPFP=0.4763 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 724,660B, BPFP=0.4600 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 857,652B, BPFP=0.5444 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 863,008B, BPFP=0.5478 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 577,996B, BPFP=0.3669 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 593,436B, BPFP=0.3767 +⌛️ [2/4] FRONTEND: Frontend time: 7.735s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.296s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14470203 4.62546908 + layer.9.1 0.14481130 4.61429621 + layer.19.0 0.11257574 13.13641153 + layer.19.1 0.11422884 12.11985091 + layer.29.0 0.10456927 157.54675821 + layer.29.1 0.10551051 144.06266250 + layer.39.0 10.36536069 3286.50731232 + layer.39.1 11.81531702 3151.27494313 + ------------------------------------------------------------------------------------- + TOTAL 2.86338443 846.73596299 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5041284 +BPFP 0.4000 bits/point +EBPFP 0.4000 equivalent bits/point +MSE 846.735963 +---------------------- --------------------------------------------------------- +Time: 21.297s Load: 1.266s, Pack+Encode: 7.735s, Decode+Unpack: 12.296s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 846.7360 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00413-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst (82/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.261s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 337,624B, BPFP=0.2143 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 347,752B, BPFP=0.2207 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 696,804B, BPFP=0.4423 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 698,076B, BPFP=0.4431 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 768,744B, BPFP=0.4880 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 767,764B, BPFP=0.4873 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 548,636B, BPFP=0.3482 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 555,768B, BPFP=0.3528 +⌛️ [2/4] FRONTEND: Frontend time: 7.707s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.305s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14530003 4.68299899 + layer.9.1 0.14546206 4.60024412 + layer.19.0 0.11891763 7.59609413 + layer.19.1 0.11677460 21.99302791 + layer.29.0 4.29725807 77.16033271 + layer.29.1 4.29692800 104.94256784 + layer.39.0 11.61914761 2630.14397140 + layer.39.1 11.22064282 2611.31134222 + ------------------------------------------------------------------------------------- + TOTAL 3.99505385 682.80382242 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4721168 +BPFP 0.3746 bits/point +EBPFP 0.3746 equivalent bits/point +MSE 682.803822 +---------------------- --------------------------------------------------------- +Time: 21.273s Load: 1.261s, Pack+Encode: 7.707s, Decode+Unpack: 12.305s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 682.8038 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00433-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst (83/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.271s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 324,412B, BPFP=0.2059 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 310,996B, BPFP=0.1974 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 626,108B, BPFP=0.3974 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 607,296B, BPFP=0.3855 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 770,428B, BPFP=0.4890 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 772,160B, BPFP=0.4901 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 571,112B, BPFP=0.3625 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 518,704B, BPFP=0.3292 +⌛️ [2/4] FRONTEND: Frontend time: 7.748s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.290s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67106792 4.76027945 + layer.9.1 2.67195307 4.72723978 + layer.19.0 0.08237472 17.50701272 + layer.19.1 0.08192194 8.10974823 + layer.29.0 0.11152953 116.43915543 + layer.29.1 0.11703055 116.81682645 + layer.39.0 163.01811830 2998.25999350 + layer.39.1 58.15221299 3105.26681833 + ------------------------------------------------------------------------------------- + TOTAL 28.36327613 796.48588424 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4501216 +BPFP 0.3571 bits/point +EBPFP 0.3571 equivalent bits/point +MSE 796.485884 +---------------------- --------------------------------------------------------- +Time: 21.309s Load: 1.271s, Pack+Encode: 7.748s, Decode+Unpack: 12.290s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 796.4859 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00443-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst (84/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.261s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 303,600B, BPFP=0.1927 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 321,564B, BPFP=0.2041 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 655,420B, BPFP=0.4160 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 681,136B, BPFP=0.4324 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 766,652B, BPFP=0.4866 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 767,844B, BPFP=0.4874 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 515,360B, BPFP=0.3271 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 504,964B, BPFP=0.3205 +⌛️ [2/4] FRONTEND: Frontend time: 7.739s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.262s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14620710 4.65048901 + layer.9.1 0.14642976 4.67947105 + layer.19.0 0.11726453 12.38414674 + layer.19.1 0.11958517 17.34331964 + layer.29.0 0.10693079 66.36374716 + layer.29.1 0.10826971 82.09790380 + layer.39.0 43.01306569 2733.73253169 + layer.39.1 17.12450997 2739.82320442 + ------------------------------------------------------------------------------------- + TOTAL 7.61028284 707.63435169 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4516540 +BPFP 0.3584 bits/point +EBPFP 0.3584 equivalent bits/point +MSE 707.634352 +---------------------- --------------------------------------------------------- +Time: 21.262s Load: 1.261s, Pack+Encode: 7.739s, Decode+Unpack: 12.262s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 707.6344 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00444-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst (85/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.265s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 313,856B, BPFP=0.1992 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 305,052B, BPFP=0.1936 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 586,216B, BPFP=0.3721 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 578,496B, BPFP=0.3672 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 729,516B, BPFP=0.4631 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 748,068B, BPFP=0.4748 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 505,148B, BPFP=0.3206 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 522,020B, BPFP=0.3314 +⌛️ [2/4] FRONTEND: Frontend time: 7.730s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.303s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03173930 4.74218750 + layer.9.1 0.03345565 4.72253723 + layer.19.0 3.26068347 8.07267821 + layer.19.1 3.26087326 7.72839502 + layer.29.0 4.24610771 47.12946864 + layer.29.1 4.24089229 47.04977454 + layer.39.0 8.81319124 2430.84156646 + layer.39.1 8.71779153 2582.81702957 + ------------------------------------------------------------------------------------- + TOTAL 4.07559181 641.63795465 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4288372 +BPFP 0.3403 bits/point +EBPFP 0.3403 equivalent bits/point +MSE 641.637955 +---------------------- --------------------------------------------------------- +Time: 21.297s Load: 1.265s, Pack+Encode: 7.730s, Decode+Unpack: 12.303s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 641.6380 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00445-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst (86/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.264s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 318,484B, BPFP=0.2022 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 322,256B, BPFP=0.2046 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 685,460B, BPFP=0.4351 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 678,020B, BPFP=0.4304 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 818,592B, BPFP=0.5196 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 814,152B, BPFP=0.5168 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 508,312B, BPFP=0.3227 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 521,040B, BPFP=0.3307 +⌛️ [2/4] FRONTEND: Frontend time: 7.744s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.295s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00078599 4.77322645 + layer.9.1 0.00079117 4.70797925 + layer.19.0 0.00795310 12.40638584 + layer.19.1 0.00811505 21.93326241 + layer.29.0 4.25797468 98.11615413 + layer.29.1 4.25504309 85.03891270 + layer.39.0 81.06806549 2803.01657459 + layer.39.1 44.82015254 2789.73415665 + ------------------------------------------------------------------------------------- + TOTAL 16.80236014 727.46583150 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4666316 +BPFP 0.3702 bits/point +EBPFP 0.3702 equivalent bits/point +MSE 727.465831 +---------------------- --------------------------------------------------------- +Time: 21.303s Load: 1.264s, Pack+Encode: 7.744s, Decode+Unpack: 12.295s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 727.4658 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00446-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst (87/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.272s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 318,312B, BPFP=0.2020 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 326,212B, BPFP=0.2071 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 757,592B, BPFP=0.4809 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 725,436B, BPFP=0.4605 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 848,608B, BPFP=0.5387 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 846,600B, BPFP=0.5374 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 546,612B, BPFP=0.3470 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 554,080B, BPFP=0.3517 +⌛️ [2/4] FRONTEND: Frontend time: 7.746s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.299s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00082351 4.72673737 + layer.9.1 0.02968625 4.74084342 + layer.19.0 0.00841222 12.39743460 + layer.19.1 0.03743129 7.30159690 + layer.29.0 4.28408194 55.87018606 + layer.29.1 4.28564945 65.96797814 + layer.39.0 8.35370986 2641.47481313 + layer.39.1 8.52557915 2614.43825154 + ------------------------------------------------------------------------------------- + TOTAL 3.19067171 675.86473015 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4923452 +BPFP 0.3906 bits/point +EBPFP 0.3906 equivalent bits/point +MSE 675.864730 +---------------------- --------------------------------------------------------- +Time: 21.318s Load: 1.272s, Pack+Encode: 7.746s, Decode+Unpack: 12.299s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 675.8647 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00447-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst (88/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.265s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 333,444B, BPFP=0.2117 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 322,840B, BPFP=0.2049 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 643,208B, BPFP=0.4083 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 711,356B, BPFP=0.4515 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 795,076B, BPFP=0.5047 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 800,584B, BPFP=0.5082 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 519,576B, BPFP=0.3298 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 520,540B, BPFP=0.3304 +⌛️ [2/4] FRONTEND: Frontend time: 7.697s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.259s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14529990 4.65509412 + layer.9.1 0.14524076 4.65640044 + layer.19.0 0.03780325 12.52960346 + layer.19.1 0.03783790 8.02692735 + layer.29.0 4.32098184 68.33776304 + layer.29.1 4.32100596 161.13992931 + layer.39.0 9.32673680 2481.36106597 + layer.39.1 9.31823369 2717.08011050 + ------------------------------------------------------------------------------------- + TOTAL 3.45664251 682.22336177 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4646624 +BPFP 0.3687 bits/point +EBPFP 0.3687 equivalent bits/point +MSE 682.223362 +---------------------- --------------------------------------------------------- +Time: 21.221s Load: 1.265s, Pack+Encode: 7.697s, Decode+Unpack: 12.259s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 682.2234 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00461-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst (89/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.262s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 328,064B, BPFP=0.2082 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 328,292B, BPFP=0.2084 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 670,972B, BPFP=0.4259 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 675,424B, BPFP=0.4287 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 800,140B, BPFP=0.5079 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 801,356B, BPFP=0.5087 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 531,304B, BPFP=0.3372 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 524,796B, BPFP=0.3331 +⌛️ [2/4] FRONTEND: Frontend time: 7.792s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.273s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14497286 4.60693597 + layer.9.1 0.14497296 4.62392727 + layer.19.0 0.03962668 26.12051867 + layer.19.1 0.11751332 21.57799297 + layer.29.0 0.14529291 120.38016940 + layer.29.1 0.16241527 131.63049439 + layer.39.0 11.40179406 2698.33246669 + layer.39.1 13.03458244 2596.07491063 + ------------------------------------------------------------------------------------- + TOTAL 3.14889631 700.41842700 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4660348 +BPFP 0.3698 bits/point +EBPFP 0.3698 equivalent bits/point +MSE 700.418427 +---------------------- --------------------------------------------------------- +Time: 21.327s Load: 1.262s, Pack+Encode: 7.792s, Decode+Unpack: 12.273s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 700.4184 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00462-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst (90/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.268s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 333,436B, BPFP=0.2116 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 344,776B, BPFP=0.2188 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 775,876B, BPFP=0.4925 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 762,572B, BPFP=0.4840 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 855,812B, BPFP=0.5432 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 852,168B, BPFP=0.5409 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 555,332B, BPFP=0.3525 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 556,124B, BPFP=0.3530 +⌛️ [2/4] FRONTEND: Frontend time: 7.719s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.287s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14513344 4.59287595 + layer.9.1 0.03283094 4.64500922 + layer.19.0 0.11544709 39.81844634 + layer.19.1 0.11326018 21.48596441 + layer.29.0 0.14483232 175.24888284 + layer.29.1 0.14672551 162.06582101 + layer.39.0 10.02784076 2966.33474163 + layer.39.1 15.62606130 2928.81507962 + ------------------------------------------------------------------------------------- + TOTAL 3.29401644 787.87585263 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5036096 +BPFP 0.3996 bits/point +EBPFP 0.3996 equivalent bits/point +MSE 787.875853 +---------------------- --------------------------------------------------------- +Time: 21.273s Load: 1.268s, Pack+Encode: 7.719s, Decode+Unpack: 12.287s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 787.8759 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00463-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst (91/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.263s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 336,404B, BPFP=0.2135 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 334,952B, BPFP=0.2126 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 793,788B, BPFP=0.5039 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 768,996B, BPFP=0.4881 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 857,576B, BPFP=0.5443 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 855,236B, BPFP=0.5429 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 589,860B, BPFP=0.3744 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 564,292B, BPFP=0.3582 +⌛️ [2/4] FRONTEND: Frontend time: 7.732s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.287s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14492303 4.59438285 + layer.9.1 0.14484742 4.59100376 + layer.19.0 0.11740684 16.84662415 + layer.19.1 0.11489933 26.21923749 + layer.29.0 0.12072669 201.35480988 + layer.29.1 0.12118037 207.54728632 + layer.39.0 10.74778980 2908.11764706 + layer.39.1 11.83662176 2830.10919727 + ------------------------------------------------------------------------------------- + TOTAL 2.91854940 774.92252360 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5101104 +BPFP 0.4047 bits/point +EBPFP 0.4047 equivalent bits/point +MSE 774.922524 +---------------------- --------------------------------------------------------- +Time: 21.282s Load: 1.263s, Pack+Encode: 7.732s, Decode+Unpack: 12.287s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 774.9225 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00464-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst (92/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.269s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 343,244B, BPFP=0.2179 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 339,268B, BPFP=0.2154 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 775,000B, BPFP=0.4919 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 770,860B, BPFP=0.4893 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 879,500B, BPFP=0.5583 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 872,472B, BPFP=0.5538 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 566,116B, BPFP=0.3593 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 554,972B, BPFP=0.3523 +⌛️ [2/4] FRONTEND: Frontend time: 7.733s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.292s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14465000 4.57190318 + layer.9.1 0.14489275 4.57273756 + layer.19.0 0.11978787 39.97639239 + layer.19.1 0.12819003 82.04425475 + layer.29.0 0.12519148 173.68093923 + layer.29.1 0.13018718 189.33498538 + layer.39.0 10.77894586 2918.65648359 + layer.39.1 10.25834823 2932.38056549 + ------------------------------------------------------------------------------------- + TOTAL 2.72877418 793.15228270 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5101432 +BPFP 0.4048 bits/point +EBPFP 0.4048 equivalent bits/point +MSE 793.152283 +---------------------- --------------------------------------------------------- +Time: 21.294s Load: 1.269s, Pack+Encode: 7.733s, Decode+Unpack: 12.292s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 793.1523 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00465-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst (93/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.260s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 294,804B, BPFP=0.1871 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 309,488B, BPFP=0.1964 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 629,012B, BPFP=0.3993 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 741,448B, BPFP=0.4706 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 756,664B, BPFP=0.4803 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 820,640B, BPFP=0.5209 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 522,152B, BPFP=0.3314 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 535,324B, BPFP=0.3398 +⌛️ [2/4] FRONTEND: Frontend time: 7.712s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.276s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14553812 4.67819362 + layer.9.1 0.14559401 4.66969590 + layer.19.0 0.04492324 12.81641260 + layer.19.1 0.04213941 7.80454786 + layer.29.0 4.25320263 47.74697859 + layer.29.1 4.25391672 71.76848899 + layer.39.0 8.72311137 2949.56223594 + layer.39.1 8.87262096 2713.02892428 + ------------------------------------------------------------------------------------- + TOTAL 3.31013081 726.50943472 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4609532 +BPFP 0.3657 bits/point +EBPFP 0.3657 equivalent bits/point +MSE 726.509435 +---------------------- --------------------------------------------------------- +Time: 21.249s Load: 1.260s, Pack+Encode: 7.712s, Decode+Unpack: 12.276s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 726.5094 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00471-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst (94/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.260s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 319,732B, BPFP=0.2029 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 324,336B, BPFP=0.2059 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 702,324B, BPFP=0.4458 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 721,592B, BPFP=0.4580 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 882,956B, BPFP=0.5605 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 881,728B, BPFP=0.5597 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 533,796B, BPFP=0.3388 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 545,000B, BPFP=0.3459 +⌛️ [2/4] FRONTEND: Frontend time: 7.701s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.314s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14518122 4.62261556 + layer.9.1 0.14529820 4.62284693 + layer.19.0 0.11833418 13.16371974 + layer.19.1 0.12038008 31.78117129 + layer.29.0 4.31360161 194.62967176 + layer.29.1 4.31792870 177.79070523 + layer.39.0 9.40764201 2686.99480013 + layer.39.1 11.30764416 2629.15014625 + ------------------------------------------------------------------------------------- + TOTAL 3.73450127 717.84445961 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4911464 +BPFP 0.3897 bits/point +EBPFP 0.3897 equivalent bits/point +MSE 717.844460 +---------------------- --------------------------------------------------------- +Time: 21.275s Load: 1.260s, Pack+Encode: 7.701s, Decode+Unpack: 12.314s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 717.8445 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00472-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst (95/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.262s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 310,728B, BPFP=0.1972 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 324,388B, BPFP=0.2059 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 772,076B, BPFP=0.4901 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 783,476B, BPFP=0.4973 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 856,608B, BPFP=0.5437 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 865,200B, BPFP=0.5492 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 540,464B, BPFP=0.3431 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 529,700B, BPFP=0.3362 +⌛️ [2/4] FRONTEND: Frontend time: 7.744s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.315s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00315138 4.66402442 + layer.9.1 0.00505826 4.65729670 + layer.19.0 0.09147678 44.89338235 + layer.19.1 0.09143778 26.22970070 + layer.29.0 0.11015094 91.50940445 + layer.29.1 0.11338039 186.49571417 + layer.39.0 9.14784464 2924.19889503 + layer.39.1 8.98944348 2872.60675983 + ------------------------------------------------------------------------------------- + TOTAL 2.31899296 769.40689721 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4982640 +BPFP 0.3953 bits/point +EBPFP 0.3953 equivalent bits/point +MSE 769.406897 +---------------------- --------------------------------------------------------- +Time: 21.320s Load: 1.262s, Pack+Encode: 7.744s, Decode+Unpack: 12.315s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 769.4069 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00475-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst (96/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.258s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 348,992B, BPFP=0.2215 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 340,596B, BPFP=0.2162 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 877,212B, BPFP=0.5568 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 786,572B, BPFP=0.4993 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 923,560B, BPFP=0.5862 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 917,080B, BPFP=0.5821 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 588,860B, BPFP=0.3738 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 566,728B, BPFP=0.3597 +⌛️ [2/4] FRONTEND: Frontend time: 7.643s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.250s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14591913 4.61954782 + layer.9.1 0.03347605 4.62172723 + layer.19.0 0.12173996 35.41241723 + layer.19.1 0.12099332 44.62147079 + layer.29.0 0.11078974 104.47662090 + layer.29.1 0.11776269 74.30442598 + layer.39.0 10.17800795 3122.50861228 + layer.39.1 9.88744998 3054.24504387 + ------------------------------------------------------------------------------------- + TOTAL 2.58951735 805.60123326 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 5349600 +BPFP 0.4245 bits/point +EBPFP 0.4245 equivalent bits/point +MSE 805.601233 +---------------------- --------------------------------------------------------- +Time: 21.151s Load: 1.258s, Pack+Encode: 7.643s, Decode+Unpack: 12.250s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 805.6012 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00476-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst (97/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.246s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 292,212B, BPFP=0.1855 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 287,796B, BPFP=0.1827 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 706,504B, BPFP=0.4485 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 700,672B, BPFP=0.4448 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 827,232B, BPFP=0.5251 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 826,748B, BPFP=0.5248 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 504,408B, BPFP=0.3202 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 519,888B, BPFP=0.3300 +⌛️ [2/4] FRONTEND: Frontend time: 7.747s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.251s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66405266 4.79792868 + layer.9.1 2.66543197 4.77236731 + layer.19.0 3.22131407 17.17158809 + layer.19.1 3.22426883 30.74635400 + layer.29.0 4.27224607 113.51440120 + layer.29.1 4.27784520 82.67683620 + layer.39.0 8.94937744 2637.42297693 + layer.39.1 8.82170070 2663.66363341 + ------------------------------------------------------------------------------------- + TOTAL 4.76202962 694.34576073 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4665460 +BPFP 0.3702 bits/point +EBPFP 0.3702 equivalent bits/point +MSE 694.345761 +---------------------- --------------------------------------------------------- +Time: 21.243s Load: 1.246s, Pack+Encode: 7.747s, Decode+Unpack: 12.251s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 694.3458 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00643-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst (98/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.254s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 316,632B, BPFP=0.2010 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 316,784B, BPFP=0.2011 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 712,768B, BPFP=0.4524 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 744,404B, BPFP=0.4725 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 839,188B, BPFP=0.5327 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 842,144B, BPFP=0.5346 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 543,300B, BPFP=0.3449 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 545,980B, BPFP=0.3466 +⌛️ [2/4] FRONTEND: Frontend time: 7.681s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.256s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00104424 4.73894837 + layer.9.1 0.00091568 4.74402161 + layer.19.0 0.08171424 7.73895916 + layer.19.1 0.08373584 7.82501917 + layer.29.0 4.26071267 112.64934392 + layer.29.1 4.26438533 85.79902299 + layer.39.0 8.39843369 2695.64348391 + layer.39.1 8.51949380 2832.06499838 + ------------------------------------------------------------------------------------- + TOTAL 3.20130444 718.90047469 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4861200 +BPFP 0.3857 bits/point +EBPFP 0.3857 equivalent bits/point +MSE 718.900475 +---------------------- --------------------------------------------------------- +Time: 21.191s Load: 1.254s, Pack+Encode: 7.681s, Decode+Unpack: 12.256s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 718.9005 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00644-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst (99/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.255s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 313,332B, BPFP=0.1989 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 316,348B, BPFP=0.2008 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 798,064B, BPFP=0.5066 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 780,120B, BPFP=0.4952 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 850,164B, BPFP=0.5396 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 844,256B, BPFP=0.5359 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 551,444B, BPFP=0.3500 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 537,248B, BPFP=0.3410 +⌛️ [2/4] FRONTEND: Frontend time: 7.689s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.260s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14622840 4.63253575 + layer.9.1 0.03344178 4.61935676 + layer.19.0 0.12675888 40.42348219 + layer.19.1 0.12382618 12.42974463 + layer.29.0 0.12223263 103.78534287 + layer.29.1 0.12797405 101.35282946 + layer.39.0 10.69978368 2826.04354891 + layer.39.1 8.63538768 2753.65745856 + ------------------------------------------------------------------------------------- + TOTAL 2.50195416 730.86803739 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4990976 +BPFP 0.3960 bits/point +EBPFP 0.3960 equivalent bits/point +MSE 730.868037 +---------------------- --------------------------------------------------------- +Time: 21.205s Load: 1.255s, Pack+Encode: 7.689s, Decode+Unpack: 12.260s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 730.8680 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00272-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst (100/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.253s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 335,432B, BPFP=0.2129 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 329,880B, BPFP=0.2094 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 700,136B, BPFP=0.4444 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 722,420B, BPFP=0.4586 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 834,144B, BPFP=0.5295 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 839,640B, BPFP=0.5330 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 524,456B, BPFP=0.3329 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 519,028B, BPFP=0.3295 +⌛️ [2/4] FRONTEND: Frontend time: 7.740s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.297s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14492994 4.60765959 + layer.9.1 0.14498602 4.60162534 + layer.19.0 0.12957112 40.59490017 + layer.19.1 0.13054295 27.27623294 + layer.29.0 0.16610158 238.63369353 + layer.29.1 0.14872770 180.50814511 + layer.39.0 16.52878844 2844.81670458 + layer.39.1 24.55764797 2721.84822879 + ------------------------------------------------------------------------------------- + TOTAL 5.24391197 757.86089876 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 4805136 +BPFP 0.3813 bits/point +EBPFP 0.3813 equivalent bits/point +MSE 757.860899 +---------------------- --------------------------------------------------------- +Time: 21.290s Load: 1.253s, Pack+Encode: 7.740s, Decode+Unpack: 12.297s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 757.8609 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst + to output-fixed/dinov3-total/lambda0.004/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00278-wflip.zst +------------------------ ---------------------------- +TOTAL PROCESSING SUMMARY +------------------------ ---------------------------- +Total files 100 +Avg BPFP 0.3721 bits/point +Avg EBPFP 0.3721 equivalent bits/point +Avg MSE 728.699276 +Avg Time 21.281s +------------------------ ---------------------------- diff --git a/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log b/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log new file mode 100644 index 0000000000000000000000000000000000000000..4b933ff29148b8bd49e05ff5da7814ecc2630f21 --- /dev/null +++ b/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log @@ -0,0 +1,15744 @@ +Experiment: dtufc_hyperprior-featurecoding_dinov3-total_individual +Log file: output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log +DTUFCCodecConfig: + arch: hyperprior-featurecoding + handler: dinov3-total + checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.007_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar + transform_type: kmeans + transform_mapping:featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json + bit_depth: 8 + device: cuda:0 +Loading checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.007_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar +Checkpoint epoch: 599 +Loaded hyperprior-featurecoding (1-channel) on cuda:0 +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_9_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.9' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_9_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_19_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.19' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_19_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_29_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.29' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_29_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_39_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.39' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_39_0.json +Loaded per-key mappings: model=dinov3-total + Keys: ['layer.9', 'layer.19', 'layer.29', 'layer.39'] +---------------- ------------------------------------------------------------------------------------------------------------------------------ +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +Checkpoint codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.007_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar +Transform type kmeans +Transform config featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json +Input ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features +Output output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test +---------------- ------------------------------------------------------------------------------------------------------------------------------ +Files found: 100 +---------------------------------------------------------------------- + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst (1/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.262s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 480,640B, BPFP=0.3051 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 480,176B, BPFP=0.3048 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,287,808B, BPFP=0.8174 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,293,140B, BPFP=0.8208 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,203,560B, BPFP=0.7640 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,213,904B, BPFP=0.7705 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 894,860B, BPFP=0.5680 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 932,128B, BPFP=0.5917 +⌛️ [2/4] FRONTEND: Frontend time: 8.151s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.029s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14561056 4.48245266 + layer.9.1 0.14522085 4.43573762 + layer.19.0 3.25142184 6.32061337 + layer.19.1 3.25206135 6.31246826 + layer.29.0 4.23946030 31.83979932 + layer.29.1 4.24539299 31.47748670 + layer.39.0 32.17105490 1579.59376016 + layer.39.1 19.15684032 1594.68199545 + ------------------------------------------------------------------------------------- + TOTAL 8.32588289 407.39303919 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7786216 +BPFP 0.6178 bits/point +EBPFP 0.6178 equivalent bits/point +MSE 407.393039 +---------------------- --------------------------------------------------------- +Time: 22.442s Load: 1.262s, Pack+Encode: 8.151s, Decode+Unpack: 13.029s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 407.3930 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00512-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst (2/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.275s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 524,596B, BPFP=0.3330 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 506,456B, BPFP=0.3215 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,288,168B, BPFP=0.8177 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,229,796B, BPFP=0.7806 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,460,920B, BPFP=0.9273 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,448,404B, BPFP=0.9194 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,084,176B, BPFP=0.6882 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,002,604B, BPFP=0.6364 +⌛️ [2/4] FRONTEND: Frontend time: 7.607s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.896s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03307773 4.43837500 + layer.9.1 0.03291117 4.40001644 + layer.19.0 0.04156009 6.37402312 + layer.19.1 0.03760627 15.28118653 + layer.29.0 4.28582750 48.99219004 + layer.29.1 4.28551552 53.58175577 + layer.39.0 9.83402183 1446.22716932 + layer.39.1 9.85397836 1494.44020149 + ------------------------------------------------------------------------------------- + TOTAL 3.55056231 384.21686471 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8545120 +BPFP 0.6780 bits/point +EBPFP 0.6780 equivalent bits/point +MSE 384.216865 +---------------------- --------------------------------------------------------- +Time: 21.778s Load: 1.275s, Pack+Encode: 7.607s, Decode+Unpack: 12.896s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 384.2169 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00657-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst (3/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.256s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 546,008B, BPFP=0.3466 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 547,548B, BPFP=0.3476 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,264,372B, BPFP=0.8026 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,272,032B, BPFP=0.8074 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,487,932B, BPFP=0.9445 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,486,548B, BPFP=0.9436 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 944,416B, BPFP=0.5995 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 953,100B, BPFP=0.6050 +⌛️ [2/4] FRONTEND: Frontend time: 7.915s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.764s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00083900 4.44332478 + layer.9.1 0.00259629 12.53093643 + layer.19.0 0.00955961 6.23601964 + layer.19.1 0.08538111 29.50337687 + layer.29.0 0.11631418 97.05769621 + layer.29.1 0.11200302 106.41693411 + layer.39.0 14.47657393 1797.44150146 + layer.39.1 13.08093694 1855.94280143 + ------------------------------------------------------------------------------------- + TOTAL 3.48552551 488.69657387 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8501956 +BPFP 0.6746 bits/point +EBPFP 0.6746 equivalent bits/point +MSE 488.696574 +---------------------- --------------------------------------------------------- +Time: 21.935s Load: 1.256s, Pack+Encode: 7.915s, Decode+Unpack: 12.764s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 488.6966 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00676-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst (4/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.247s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 483,596B, BPFP=0.3070 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 493,804B, BPFP=0.3134 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,360,868B, BPFP=0.8638 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,334,808B, BPFP=0.8473 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,281,212B, BPFP=0.8132 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,287,076B, BPFP=0.8170 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 926,456B, BPFP=0.5881 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 943,948B, BPFP=0.5992 +⌛️ [2/4] FRONTEND: Frontend time: 7.668s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.037s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.70169554 4.45827853 + layer.9.1 0.03294074 4.45034416 + layer.19.0 3.25671692 6.74554152 + layer.19.1 3.25834093 24.92178106 + layer.29.0 0.10810242 119.11360497 + layer.29.1 0.10661203 153.63301308 + layer.39.0 8.95005916 1356.76535587 + layer.39.1 8.98756017 1316.32442314 + ------------------------------------------------------------------------------------- + TOTAL 3.42525349 373.30154279 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8111768 +BPFP 0.6436 bits/point +EBPFP 0.6436 equivalent bits/point +MSE 373.301543 +---------------------- --------------------------------------------------------- +Time: 21.952s Load: 1.247s, Pack+Encode: 7.668s, Decode+Unpack: 13.037s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 373.3015 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00688-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst (5/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.278s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 517,880B, BPFP=0.3287 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 518,812B, BPFP=0.3293 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,384,156B, BPFP=0.8786 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,368,392B, BPFP=0.8686 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,403,384B, BPFP=0.8908 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,412,328B, BPFP=0.8965 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 937,616B, BPFP=0.5952 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 931,452B, BPFP=0.5912 +⌛️ [2/4] FRONTEND: Frontend time: 7.882s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.886s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14527870 4.41112520 + layer.9.1 0.14521496 4.40121802 + layer.19.0 0.03964342 6.06959523 + layer.19.1 0.03956446 15.33704704 + layer.29.0 0.12258449 159.39593151 + layer.29.1 0.12735008 134.64397140 + layer.39.0 32.94776263 1721.21660708 + layer.39.1 29.25669534 1641.72895678 + ------------------------------------------------------------------------------------- + TOTAL 7.85301176 460.90055653 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8474020 +BPFP 0.6724 bits/point +EBPFP 0.6724 equivalent bits/point +MSE 460.900557 +---------------------- --------------------------------------------------------- +Time: 22.046s Load: 1.278s, Pack+Encode: 7.882s, Decode+Unpack: 12.886s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 460.9006 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00693-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst (6/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.250s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 477,880B, BPFP=0.3033 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 492,068B, BPFP=0.3123 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,356,904B, BPFP=0.8613 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,360,592B, BPFP=0.8636 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,270,352B, BPFP=0.8064 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,262,180B, BPFP=0.8012 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 895,500B, BPFP=0.5684 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 900,824B, BPFP=0.5718 +⌛️ [2/4] FRONTEND: Frontend time: 7.961s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.070s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66816044 4.50698955 + layer.9.1 2.66817504 4.49558119 + layer.19.0 3.22262959 6.71156273 + layer.19.1 3.22037432 6.48514241 + layer.29.0 4.30448692 140.91160221 + layer.29.1 4.31085282 104.91793955 + layer.39.0 38.33931691 1423.73610660 + layer.39.1 57.25219370 1540.37130322 + ------------------------------------------------------------------------------------- + TOTAL 14.49827372 404.01702843 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8016300 +BPFP 0.6360 bits/point +EBPFP 0.6360 equivalent bits/point +MSE 404.017028 +---------------------- --------------------------------------------------------- +Time: 22.281s Load: 1.250s, Pack+Encode: 7.961s, Decode+Unpack: 13.070s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 404.0170 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00731-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst (7/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.273s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 518,176B, BPFP=0.3289 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 503,664B, BPFP=0.3197 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,211,456B, BPFP=0.7690 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,241,708B, BPFP=0.7882 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,326,816B, BPFP=0.8422 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,327,404B, BPFP=0.8426 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 903,516B, BPFP=0.5735 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 900,640B, BPFP=0.5717 +⌛️ [2/4] FRONTEND: Frontend time: 7.616s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.642s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11332920 4.49305108 + layer.9.1 0.00092169 4.50159417 + layer.19.0 3.23006092 6.41907004 + layer.19.1 3.23257961 38.92849163 + layer.29.0 4.28548854 62.05972741 + layer.29.1 4.27808990 85.31963967 + layer.39.0 10.57841825 1451.83295418 + layer.39.1 20.33118703 1471.19223269 + ------------------------------------------------------------------------------------- + TOTAL 5.75625939 390.59334511 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7933380 +BPFP 0.6295 bits/point +EBPFP 0.6295 equivalent bits/point +MSE 390.593345 +---------------------- --------------------------------------------------------- +Time: 21.531s Load: 1.273s, Pack+Encode: 7.616s, Decode+Unpack: 12.642s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 390.5933 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00061-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst (8/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.255s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 524,048B, BPFP=0.3326 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 526,804B, BPFP=0.3344 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,232,112B, BPFP=0.7821 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,336,768B, BPFP=0.8485 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,289,520B, BPFP=0.8185 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,327,704B, BPFP=0.8428 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 906,756B, BPFP=0.5756 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 889,928B, BPFP=0.5649 +⌛️ [2/4] FRONTEND: Frontend time: 7.683s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.035s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14443954 12.48275385 + layer.9.1 0.14435121 4.37580867 + layer.19.0 0.03807715 6.53925102 + layer.19.1 0.03781311 20.16435195 + layer.29.0 0.10781899 59.01390356 + layer.29.1 0.10618912 53.44404046 + layer.39.0 9.30898666 1450.57653559 + layer.39.1 9.83625107 1487.28339292 + ------------------------------------------------------------------------------------- + TOTAL 2.46549086 386.73500475 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8033640 +BPFP 0.6374 bits/point +EBPFP 0.6374 equivalent bits/point +MSE 386.735005 +---------------------- --------------------------------------------------------- +Time: 21.974s Load: 1.255s, Pack+Encode: 7.683s, Decode+Unpack: 13.035s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 386.7350 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00175-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst (9/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.278s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 535,376B, BPFP=0.3398 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 531,736B, BPFP=0.3375 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,316,228B, BPFP=0.8355 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,352,300B, BPFP=0.8584 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,482,932B, BPFP=0.9413 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,474,052B, BPFP=0.9357 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 977,152B, BPFP=0.6202 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 968,712B, BPFP=0.6149 +⌛️ [2/4] FRONTEND: Frontend time: 7.972s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.611s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14543928 4.44246819 + layer.9.1 0.14562574 4.56304557 + layer.19.0 0.11552505 16.19621638 + layer.19.1 0.12052174 6.39642344 + layer.29.0 0.10841144 49.55183113 + layer.29.1 0.10845811 53.48724407 + layer.39.0 9.17501701 1527.78940526 + layer.39.1 9.20635778 1509.40380240 + ------------------------------------------------------------------------------------- + TOTAL 2.39066952 396.47880456 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8638488 +BPFP 0.6854 bits/point +EBPFP 0.6854 equivalent bits/point +MSE 396.478805 +---------------------- --------------------------------------------------------- +Time: 21.861s Load: 1.278s, Pack+Encode: 7.972s, Decode+Unpack: 12.611s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 396.4788 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00186-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst (10/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.270s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 499,908B, BPFP=0.3173 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 488,328B, BPFP=0.3100 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,276,072B, BPFP=0.8100 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,209,716B, BPFP=0.7679 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,172,828B, BPFP=0.7445 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,171,568B, BPFP=0.7437 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 913,988B, BPFP=0.5802 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 909,620B, BPFP=0.5774 +⌛️ [2/4] FRONTEND: Frontend time: 7.817s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.126s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.78226592 8.49295237 + layer.9.1 2.78427046 4.46407697 + layer.19.0 3.22580366 6.91843402 + layer.19.1 3.22969594 10.99154132 + layer.29.0 4.29525448 96.64513934 + layer.29.1 0.11349234 81.75190933 + layer.39.0 8.89338553 1347.51226844 + layer.39.1 8.88767087 1317.09392265 + ------------------------------------------------------------------------------------- + TOTAL 4.27647990 359.23378056 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7642028 +BPFP 0.6063 bits/point +EBPFP 0.6063 equivalent bits/point +MSE 359.233781 +---------------------- --------------------------------------------------------- +Time: 22.213s Load: 1.270s, Pack+Encode: 7.817s, Decode+Unpack: 13.126s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 359.2338 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00282-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst (11/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.271s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 501,308B, BPFP=0.3182 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 487,052B, BPFP=0.3092 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,234,612B, BPFP=0.7837 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,201,352B, BPFP=0.7626 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,234,732B, BPFP=0.7837 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,232,480B, BPFP=0.7823 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 873,360B, BPFP=0.5544 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 862,632B, BPFP=0.5476 +⌛️ [2/4] FRONTEND: Frontend time: 7.821s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.952s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14514284 16.99417681 + layer.9.1 0.14518188 8.77440676 + layer.19.0 0.04057091 20.78790345 + layer.19.1 0.04041447 6.65194005 + layer.29.0 4.25641542 40.87605368 + layer.29.1 4.26613502 43.49367281 + layer.39.0 12.58558458 1474.61082223 + layer.39.1 8.96866240 1479.82011700 + ------------------------------------------------------------------------------------- + TOTAL 3.80601344 386.50113660 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7627528 +BPFP 0.6052 bits/point +EBPFP 0.6052 equivalent bits/point +MSE 386.501137 +---------------------- --------------------------------------------------------- +Time: 22.044s Load: 1.271s, Pack+Encode: 7.821s, Decode+Unpack: 12.952s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 386.5011 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00524-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst (12/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.281s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 488,248B, BPFP=0.3099 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 493,564B, BPFP=0.3133 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,262,156B, BPFP=0.8012 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,324,812B, BPFP=0.8409 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,276,124B, BPFP=0.8100 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,294,040B, BPFP=0.8214 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 867,996B, BPFP=0.5510 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 898,376B, BPFP=0.5702 +⌛️ [2/4] FRONTEND: Frontend time: 8.023s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.747s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67015388 4.48298998 + layer.9.1 0.00076871 4.45818935 + layer.19.0 3.22151687 11.13783845 + layer.19.1 3.22388957 6.42044554 + layer.29.0 4.24084786 33.84011923 + layer.29.1 4.24602234 48.31621202 + layer.39.0 7.87160790 1305.89957751 + layer.39.1 9.85764150 1223.09034774 + ------------------------------------------------------------------------------------- + TOTAL 4.41655608 329.70571498 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7905316 +BPFP 0.6272 bits/point +EBPFP 0.6272 equivalent bits/point +MSE 329.705715 +---------------------- --------------------------------------------------------- +Time: 22.051s Load: 1.281s, Pack+Encode: 8.023s, Decode+Unpack: 12.747s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 329.7057 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00538-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst (13/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.251s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 532,420B, BPFP=0.3380 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 531,972B, BPFP=0.3377 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,228,696B, BPFP=0.7799 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,237,456B, BPFP=0.7855 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,443,228B, BPFP=0.9161 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,438,296B, BPFP=0.9130 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 953,056B, BPFP=0.6050 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 943,044B, BPFP=0.5986 +⌛️ [2/4] FRONTEND: Frontend time: 7.963s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.196s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00068237 4.45202244 + layer.9.1 0.00070576 4.44733926 + layer.19.0 0.00823322 25.11928217 + layer.19.1 0.08594799 15.90481293 + layer.29.0 0.12200666 156.42293630 + layer.29.1 0.12451052 102.23848310 + layer.39.0 55.99513528 1964.51104972 + layer.39.1 28.81185256 2038.97253819 + ------------------------------------------------------------------------------------- + TOTAL 10.64363429 539.00855801 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8308168 +BPFP 0.6592 bits/point +EBPFP 0.6592 equivalent bits/point +MSE 539.008558 +---------------------- --------------------------------------------------------- +Time: 22.410s Load: 1.251s, Pack+Encode: 7.963s, Decode+Unpack: 13.196s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 539.0086 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00927-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst (14/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.277s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 523,684B, BPFP=0.3324 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 533,428B, BPFP=0.3386 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,227,748B, BPFP=0.7793 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,295,396B, BPFP=0.8223 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,263,080B, BPFP=0.8017 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,286,628B, BPFP=0.8167 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 912,796B, BPFP=0.5794 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 920,108B, BPFP=0.5840 +⌛️ [2/4] FRONTEND: Frontend time: 7.624s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.911s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14556937 8.51739278 + layer.9.1 0.03327741 4.56495331 + layer.19.0 0.11590617 15.89408312 + layer.19.1 0.11733878 34.56472416 + layer.29.0 0.11334742 87.10552080 + layer.29.1 4.29039579 53.11000467 + layer.39.0 9.10722066 1389.46148846 + layer.39.1 44.52401893 1440.01202470 + ------------------------------------------------------------------------------------- + TOTAL 7.30588432 379.15377400 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7962868 +BPFP 0.6318 bits/point +EBPFP 0.6318 equivalent bits/point +MSE 379.153774 +---------------------- --------------------------------------------------------- +Time: 21.813s Load: 1.277s, Pack+Encode: 7.624s, Decode+Unpack: 12.911s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 379.1538 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00933-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst (15/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.277s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 552,360B, BPFP=0.3506 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 552,712B, BPFP=0.3508 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,283,584B, BPFP=0.8148 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,319,912B, BPFP=0.8378 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,366,472B, BPFP=0.8674 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,380,556B, BPFP=0.8763 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 842,964B, BPFP=0.5351 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 862,820B, BPFP=0.5477 +⌛️ [2/4] FRONTEND: Frontend time: 7.815s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.947s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11318631 17.14124959 + layer.9.1 0.11319129 16.96182869 + layer.19.0 0.00665199 6.24984195 + layer.19.1 0.00853768 6.43480485 + layer.29.0 4.27225940 37.56021490 + layer.29.1 4.27324961 32.92598818 + layer.39.0 14.80262837 1542.06548586 + layer.39.1 16.56649765 1510.08547286 + ------------------------------------------------------------------------------------- + TOTAL 5.01952529 396.17811086 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8161380 +BPFP 0.6476 bits/point +EBPFP 0.6476 equivalent bits/point +MSE 396.178111 +---------------------- --------------------------------------------------------- +Time: 22.039s Load: 1.277s, Pack+Encode: 7.815s, Decode+Unpack: 12.947s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 396.1781 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01095-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst (16/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.251s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 524,380B, BPFP=0.3329 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 523,520B, BPFP=0.3323 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,280,772B, BPFP=0.8130 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,247,244B, BPFP=0.7917 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,349,500B, BPFP=0.8566 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,342,656B, BPFP=0.8523 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 904,568B, BPFP=0.5742 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 892,404B, BPFP=0.5665 +⌛️ [2/4] FRONTEND: Frontend time: 8.052s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.777s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00065117 4.46210671 + layer.9.1 0.00066201 4.45323227 + layer.19.0 0.00984582 6.18393334 + layer.19.1 0.01156107 6.48758366 + layer.29.0 4.26547583 56.62278599 + layer.29.1 4.26296603 28.26558692 + layer.39.0 11.21169412 1414.17663308 + layer.39.1 9.31977106 1412.53298668 + ------------------------------------------------------------------------------------- + TOTAL 3.63532839 366.64810608 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8065044 +BPFP 0.6399 bits/point +EBPFP 0.6399 equivalent bits/point +MSE 366.648106 +---------------------- --------------------------------------------------------- +Time: 22.081s Load: 1.251s, Pack+Encode: 8.052s, Decode+Unpack: 12.777s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 366.6481 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01125-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst (17/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.252s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 500,440B, BPFP=0.3177 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 496,308B, BPFP=0.3150 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,209,388B, BPFP=0.7677 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,171,764B, BPFP=0.7438 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,228,672B, BPFP=0.7799 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,234,576B, BPFP=0.7836 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 874,048B, BPFP=0.5548 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 871,520B, BPFP=0.5532 +⌛️ [2/4] FRONTEND: Frontend time: 7.996s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.093s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00085355 4.49703445 + layer.9.1 0.00085581 4.50012251 + layer.19.0 0.00808159 6.61499634 + layer.19.1 0.00635426 6.77066047 + layer.29.0 4.24551200 39.64853144 + layer.29.1 4.24803037 34.73965358 + layer.39.0 9.19283951 1380.63958401 + layer.39.1 9.46657027 1358.89909002 + ------------------------------------------------------------------------------------- + TOTAL 3.39613717 354.53870910 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7586716 +BPFP 0.6020 bits/point +EBPFP 0.6020 equivalent bits/point +MSE 354.538709 +---------------------- --------------------------------------------------------- +Time: 22.341s Load: 1.252s, Pack+Encode: 7.996s, Decode+Unpack: 13.093s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 354.5387 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01135-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst (18/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.268s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 517,344B, BPFP=0.3284 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 520,496B, BPFP=0.3304 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,274,088B, BPFP=0.8087 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,264,444B, BPFP=0.8026 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,407,596B, BPFP=0.8935 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,413,188B, BPFP=0.8970 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 943,936B, BPFP=0.5992 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 929,848B, BPFP=0.5902 +⌛️ [2/4] FRONTEND: Frontend time: 7.553s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.733s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67056234 4.49269562 + layer.9.1 2.67147828 4.46854688 + layer.19.0 0.00618387 11.10749741 + layer.19.1 0.08383032 25.30899872 + layer.29.0 4.28489822 33.63454410 + layer.29.1 4.28470970 33.34209965 + layer.39.0 10.15376305 1453.10042249 + layer.39.1 8.47863686 1424.77283068 + ------------------------------------------------------------------------------------- + TOTAL 4.07925783 373.77845444 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8270940 +BPFP 0.6562 bits/point +EBPFP 0.6562 equivalent bits/point +MSE 373.778454 +---------------------- --------------------------------------------------------- +Time: 21.554s Load: 1.268s, Pack+Encode: 7.553s, Decode+Unpack: 12.733s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 373.7785 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01161-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst (19/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.269s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 518,460B, BPFP=0.3291 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 511,040B, BPFP=0.3244 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,309,620B, BPFP=0.8313 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,302,072B, BPFP=0.8265 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,379,072B, BPFP=0.8754 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,385,208B, BPFP=0.8793 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 899,272B, BPFP=0.5708 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 891,640B, BPFP=0.5660 +⌛️ [2/4] FRONTEND: Frontend time: 7.540s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.613s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67132404 4.49390450 + layer.9.1 2.67117709 4.47124457 + layer.19.0 0.00597838 15.28015061 + layer.19.1 0.00605309 15.55073682 + layer.29.0 4.29273040 96.44438577 + layer.29.1 4.29206328 46.75581939 + layer.39.0 9.96127074 1512.83344166 + layer.39.1 10.21295854 1488.40591485 + ------------------------------------------------------------------------------------- + TOTAL 4.26419445 398.02944977 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8196384 +BPFP 0.6503 bits/point +EBPFP 0.6503 equivalent bits/point +MSE 398.029450 +---------------------- --------------------------------------------------------- +Time: 21.422s Load: 1.269s, Pack+Encode: 7.540s, Decode+Unpack: 12.613s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 398.0294 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01162-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst (20/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.160s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 497,808B, BPFP=0.3160 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 508,692B, BPFP=0.3229 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,294,328B, BPFP=0.8216 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,265,456B, BPFP=0.8032 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,203,996B, BPFP=0.7642 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,230,608B, BPFP=0.7811 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 862,888B, BPFP=0.5477 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 865,960B, BPFP=0.5497 +⌛️ [2/4] FRONTEND: Frontend time: 7.605s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.080s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14568686 4.43776025 + layer.9.1 0.14558674 4.43515238 + layer.19.0 0.00960369 6.32381441 + layer.19.1 0.03847206 10.91405488 + layer.29.0 4.24438723 27.33629804 + layer.29.1 4.24578970 33.52062429 + layer.39.0 9.23757985 1393.13194670 + layer.39.1 9.43674592 1447.63795905 + ------------------------------------------------------------------------------------- + TOTAL 3.43798151 365.96720125 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7729736 +BPFP 0.6133 bits/point +EBPFP 0.6133 equivalent bits/point +MSE 365.967201 +---------------------- --------------------------------------------------------- +Time: 21.846s Load: 1.160s, Pack+Encode: 7.605s, Decode+Unpack: 13.080s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 365.9672 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01163-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst (21/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.215s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 534,340B, BPFP=0.3392 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 536,596B, BPFP=0.3406 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,326,872B, BPFP=0.8422 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,333,820B, BPFP=0.8466 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,465,728B, BPFP=0.9304 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,450,404B, BPFP=0.9206 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 911,440B, BPFP=0.5785 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 908,680B, BPFP=0.5768 +⌛️ [2/4] FRONTEND: Frontend time: 7.672s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.191s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11307967 8.47054825 + layer.9.1 0.00073224 4.45397303 + layer.19.0 0.08207503 11.55141727 + layer.19.1 0.08214869 29.65204542 + layer.29.0 4.26728487 56.84113991 + layer.29.1 4.26774951 27.01651365 + layer.39.0 12.81553410 1552.35147871 + layer.39.1 23.05196315 1492.74065648 + ------------------------------------------------------------------------------------- + TOTAL 5.58507091 397.88472159 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8467880 +BPFP 0.6719 bits/point +EBPFP 0.6719 equivalent bits/point +MSE 397.884722 +---------------------- --------------------------------------------------------- +Time: 22.079s Load: 1.215s, Pack+Encode: 7.672s, Decode+Unpack: 13.191s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 397.8847 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01195-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst (22/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.271s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 639,976B, BPFP=0.4062 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 636,516B, BPFP=0.4040 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,381,380B, BPFP=0.8768 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,388,224B, BPFP=0.8812 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,674,648B, BPFP=1.0630 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,680,124B, BPFP=1.0665 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 944,208B, BPFP=0.5993 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 972,568B, BPFP=0.6173 +⌛️ [2/4] FRONTEND: Frontend time: 7.680s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.055s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14519844 54.66635014 + layer.9.1 0.14499054 29.39035079 + layer.19.0 0.12156012 194.05516737 + layer.19.1 0.12030756 165.99965470 + layer.29.0 0.12020218 58.77313028 + layer.29.1 0.12115470 99.13940120 + layer.39.0 8.85439666 1720.11407215 + layer.39.1 8.75438231 1645.95287618 + ------------------------------------------------------------------------------------- + TOTAL 2.29777406 496.01137535 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9317644 +BPFP 0.7393 bits/point +EBPFP 0.7393 equivalent bits/point +MSE 496.011375 +---------------------- --------------------------------------------------------- +Time: 22.006s Load: 1.271s, Pack+Encode: 7.680s, Decode+Unpack: 13.055s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 496.0114 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00083-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst (23/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.259s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 635,440B, BPFP=0.4033 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 639,336B, BPFP=0.4058 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,383,700B, BPFP=0.8783 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,379,996B, BPFP=0.8760 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,636,992B, BPFP=1.0391 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,621,844B, BPFP=1.0295 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 861,564B, BPFP=0.5469 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 851,652B, BPFP=0.5406 +⌛️ [2/4] FRONTEND: Frontend time: 8.099s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.237s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14541288 70.80954461 + layer.9.1 0.14479464 62.50782519 + layer.19.0 0.11855170 129.39096523 + layer.19.1 0.11778439 192.29854566 + layer.29.0 0.12648388 79.09640579 + layer.29.1 0.12520221 73.04923627 + layer.39.0 8.37129624 1596.29184270 + layer.39.1 8.45478741 1656.14575886 + ------------------------------------------------------------------------------------- + TOTAL 2.20053917 482.44876554 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9010524 +BPFP 0.7149 bits/point +EBPFP 0.7149 equivalent bits/point +MSE 482.448766 +---------------------- --------------------------------------------------------- +Time: 22.595s Load: 1.259s, Pack+Encode: 8.099s, Decode+Unpack: 13.237s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 482.4488 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00084-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst (24/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.278s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 646,840B, BPFP=0.4106 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 657,184B, BPFP=0.4171 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,467,708B, BPFP=0.9316 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,466,044B, BPFP=0.9306 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,785,356B, BPFP=1.1333 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,786,768B, BPFP=1.1342 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,239,320B, BPFP=0.7867 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,324,612B, BPFP=0.8408 +⌛️ [2/4] FRONTEND: Frontend time: 7.521s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.638s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14458197 8.81297289 + layer.9.1 0.14461228 20.79286846 + layer.19.0 0.12127609 119.71181345 + layer.19.1 0.12505172 77.29655509 + layer.29.0 0.11568762 49.07648481 + layer.29.1 0.11796058 44.44716343 + layer.39.0 8.63782956 1665.92476438 + layer.39.1 8.69862780 1748.80809230 + ------------------------------------------------------------------------------------- + TOTAL 2.26320345 466.85883935 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10373832 +BPFP 0.8231 bits/point +EBPFP 0.8231 equivalent bits/point +MSE 466.858839 +---------------------- --------------------------------------------------------- +Time: 21.437s Load: 1.278s, Pack+Encode: 7.521s, Decode+Unpack: 12.638s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 466.8588 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00085-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst (25/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.281s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 646,180B, BPFP=0.4102 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 651,904B, BPFP=0.4138 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,480,740B, BPFP=0.9399 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,480,648B, BPFP=0.9398 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,789,156B, BPFP=1.1357 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,789,988B, BPFP=1.1362 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,305,272B, BPFP=0.8285 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,308,016B, BPFP=0.8303 +⌛️ [2/4] FRONTEND: Frontend time: 7.696s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.223s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14478087 8.73464477 + layer.9.1 0.14472154 9.01243030 + layer.19.0 0.13423899 169.54032946 + layer.19.1 0.13534726 110.60101154 + layer.29.0 0.11251127 68.44804192 + layer.29.1 0.11242151 68.48639604 + layer.39.0 10.58490794 1775.18833279 + layer.39.1 8.80008176 1762.84546636 + ------------------------------------------------------------------------------------- + TOTAL 2.52112639 496.60708165 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10451904 +BPFP 0.8293 bits/point +EBPFP 0.8293 equivalent bits/point +MSE 496.607082 +---------------------- --------------------------------------------------------- +Time: 22.200s Load: 1.281s, Pack+Encode: 7.696s, Decode+Unpack: 13.223s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 496.6071 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00086-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst (26/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.280s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 608,232B, BPFP=0.3861 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 615,748B, BPFP=0.3908 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,329,996B, BPFP=0.8442 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,349,532B, BPFP=0.8566 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,618,628B, BPFP=1.0274 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,612,884B, BPFP=1.0238 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,064,712B, BPFP=0.6758 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,096,276B, BPFP=0.6959 +⌛️ [2/4] FRONTEND: Frontend time: 7.635s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.131s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14590075 24.91916589 + layer.9.1 0.14620647 33.56426714 + layer.19.0 0.11628058 20.86543051 + layer.19.1 0.11601873 89.77294240 + layer.29.0 0.11558260 46.25544361 + layer.29.1 0.11828149 56.14544402 + layer.39.0 28.43028163 1948.70328242 + layer.39.1 24.81181701 1980.91403965 + ------------------------------------------------------------------------------------- + TOTAL 6.75004616 525.14250196 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9296008 +BPFP 0.7376 bits/point +EBPFP 0.7376 equivalent bits/point +MSE 525.142502 +---------------------- --------------------------------------------------------- +Time: 22.047s Load: 1.280s, Pack+Encode: 7.635s, Decode+Unpack: 13.131s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 525.1425 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00090-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst (27/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.276s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 593,564B, BPFP=0.3768 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 579,400B, BPFP=0.3678 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,345,624B, BPFP=0.8541 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,339,616B, BPFP=0.8503 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,677,116B, BPFP=1.0645 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,661,960B, BPFP=1.0549 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,220,840B, BPFP=0.7749 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,132,452B, BPFP=0.7188 +⌛️ [2/4] FRONTEND: Frontend time: 7.586s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.798s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11529700 4.72048763 + layer.9.1 0.14629077 4.45077643 + layer.19.0 0.09721754 39.14627173 + layer.19.1 0.12446257 62.33348737 + layer.29.0 4.28687864 46.61276710 + layer.29.1 4.28715508 86.00538268 + layer.39.0 11.34089363 1800.36545336 + layer.39.1 19.75513766 1823.47416315 + ------------------------------------------------------------------------------------- + TOTAL 5.01916661 483.38859868 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9550572 +BPFP 0.7578 bits/point +EBPFP 0.7578 equivalent bits/point +MSE 483.388599 +---------------------- --------------------------------------------------------- +Time: 21.660s Load: 1.276s, Pack+Encode: 7.586s, Decode+Unpack: 12.798s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 483.3886 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00116-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst (28/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.263s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 577,592B, BPFP=0.3666 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 584,900B, BPFP=0.3713 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,302,980B, BPFP=0.8271 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,288,404B, BPFP=0.8178 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,549,688B, BPFP=0.9837 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,557,808B, BPFP=0.9888 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,081,856B, BPFP=0.6867 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,119,492B, BPFP=0.7106 +⌛️ [2/4] FRONTEND: Frontend time: 7.542s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.941s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14538718 4.44000694 + layer.9.1 0.14538559 4.42265987 + layer.19.0 0.11434236 43.88064673 + layer.19.1 0.11406084 76.88263731 + layer.29.0 0.11219077 89.39212301 + layer.29.1 0.11281304 65.07120877 + layer.39.0 79.88316542 1976.09684758 + layer.39.1 46.71980622 1985.79996750 + ------------------------------------------------------------------------------------- + TOTAL 15.91839393 530.74826221 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9062720 +BPFP 0.7191 bits/point +EBPFP 0.7191 equivalent bits/point +MSE 530.748262 +---------------------- --------------------------------------------------------- +Time: 21.745s Load: 1.263s, Pack+Encode: 7.542s, Decode+Unpack: 12.941s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 530.7483 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00314-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst (29/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.248s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 614,152B, BPFP=0.3898 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 609,856B, BPFP=0.3871 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,323,976B, BPFP=0.8404 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,332,592B, BPFP=0.8459 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,606,628B, BPFP=1.0198 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,598,296B, BPFP=1.0145 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,228,484B, BPFP=0.7798 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,191,968B, BPFP=0.7566 +⌛️ [2/4] FRONTEND: Frontend time: 7.835s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.262s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14512126 12.96325053 + layer.9.1 0.14517278 8.62497398 + layer.19.0 0.11689420 126.88022018 + layer.19.1 0.12099910 129.45090591 + layer.29.0 0.11847120 66.00436708 + layer.29.1 0.12399357 66.07518992 + layer.39.0 75.86630139 2015.05411115 + layer.39.1 56.61936342 1981.00536237 + ------------------------------------------------------------------------------------- + TOTAL 16.65703962 550.75729764 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9505952 +BPFP 0.7542 bits/point +EBPFP 0.7542 equivalent bits/point +MSE 550.757298 +---------------------- --------------------------------------------------------- +Time: 22.345s Load: 1.248s, Pack+Encode: 7.835s, Decode+Unpack: 13.262s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 550.7573 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00316-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst (30/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.275s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 591,980B, BPFP=0.3758 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 584,080B, BPFP=0.3707 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,324,744B, BPFP=0.8409 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,316,208B, BPFP=0.8355 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,566,804B, BPFP=0.9945 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,563,700B, BPFP=0.9926 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,178,564B, BPFP=0.7481 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,180,416B, BPFP=0.7493 +⌛️ [2/4] FRONTEND: Frontend time: 7.783s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.102s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11488649 4.58884275 + layer.9.1 0.14606862 8.45439133 + layer.19.0 0.08767178 76.15751950 + layer.19.1 0.11443626 35.47419869 + layer.29.0 0.10933029 83.15305594 + layer.29.1 0.10817130 49.41229790 + layer.39.0 52.66717785 1950.74049399 + layer.39.1 62.91127214 2024.67468313 + ------------------------------------------------------------------------------------- + TOTAL 14.53237684 529.08193540 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9306496 +BPFP 0.7384 bits/point +EBPFP 0.7384 equivalent bits/point +MSE 529.081935 +---------------------- --------------------------------------------------------- +Time: 22.161s Load: 1.275s, Pack+Encode: 7.783s, Decode+Unpack: 13.102s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 529.0819 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00324-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst (31/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.234s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 597,692B, BPFP=0.3794 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 594,788B, BPFP=0.3775 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,279,424B, BPFP=0.8121 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,289,220B, BPFP=0.8183 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,532,100B, BPFP=0.9725 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,534,040B, BPFP=0.9737 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,088,548B, BPFP=0.6910 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,118,856B, BPFP=0.7102 +⌛️ [2/4] FRONTEND: Frontend time: 7.661s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.490s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14509552 4.54629071 + layer.9.1 0.14520687 4.42504780 + layer.19.0 0.12118574 20.39400695 + layer.19.1 0.11709642 34.57023887 + layer.29.0 0.10963326 64.53073204 + layer.29.1 0.10842036 59.91913288 + layer.39.0 53.79489966 1905.80614235 + layer.39.1 62.27410526 1817.03802405 + ------------------------------------------------------------------------------------- + TOTAL 14.60195539 488.90370196 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9034668 +BPFP 0.7168 bits/point +EBPFP 0.7168 equivalent bits/point +MSE 488.903702 +---------------------- --------------------------------------------------------- +Time: 21.385s Load: 1.234s, Pack+Encode: 7.661s, Decode+Unpack: 12.490s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 488.9037 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00325-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst (32/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.228s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 668,692B, BPFP=0.4245 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 659,060B, BPFP=0.4183 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,380,768B, BPFP=0.8764 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,383,960B, BPFP=0.8785 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,641,200B, BPFP=1.0418 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,637,876B, BPFP=1.0396 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,195,276B, BPFP=0.7587 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,185,236B, BPFP=0.7523 +⌛️ [2/4] FRONTEND: Frontend time: 7.882s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.300s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14577075 17.20370262 + layer.9.1 0.14541274 20.44881124 + layer.19.0 0.13069581 85.37398440 + layer.19.1 0.13545482 123.20576048 + layer.29.0 0.11331055 64.55964617 + layer.29.1 0.11244963 89.03926308 + layer.39.0 32.27446072 1916.80711732 + layer.39.1 16.59366367 1820.83392915 + ------------------------------------------------------------------------------------- + TOTAL 6.20640234 517.18402681 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9752068 +BPFP 0.7738 bits/point +EBPFP 0.7738 equivalent bits/point +MSE 517.184027 +---------------------- --------------------------------------------------------- +Time: 22.410s Load: 1.228s, Pack+Encode: 7.882s, Decode+Unpack: 13.300s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 517.1840 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00327-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst (33/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.274s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 567,592B, BPFP=0.3603 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 566,220B, BPFP=0.3594 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,287,176B, BPFP=0.8170 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,302,496B, BPFP=0.8268 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,502,172B, BPFP=0.9535 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,493,372B, BPFP=0.9479 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,118,356B, BPFP=0.7099 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,050,096B, BPFP=0.6665 +⌛️ [2/4] FRONTEND: Frontend time: 7.797s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.804s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14585920 4.44520586 + layer.9.1 0.14576220 32.87766849 + layer.19.0 0.12270736 68.43200561 + layer.19.1 0.12453605 53.70123091 + layer.29.0 0.11393550 114.63858872 + layer.29.1 0.11678154 85.77328770 + layer.39.0 53.83016636 1785.55427364 + layer.39.1 40.65720720 1709.31475463 + ------------------------------------------------------------------------------------- + TOTAL 11.90711942 481.84212694 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8887480 +BPFP 0.7052 bits/point +EBPFP 0.7052 equivalent bits/point +MSE 481.842127 +---------------------- --------------------------------------------------------- +Time: 21.875s Load: 1.274s, Pack+Encode: 7.797s, Decode+Unpack: 12.804s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 481.8421 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00328-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst (34/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.280s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 563,928B, BPFP=0.3580 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 561,764B, BPFP=0.3566 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,345,388B, BPFP=0.8540 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,346,992B, BPFP=0.8550 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,548,352B, BPFP=0.9828 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,551,912B, BPFP=0.9851 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 996,724B, BPFP=0.6327 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,006,348B, BPFP=0.6388 +⌛️ [2/4] FRONTEND: Frontend time: 7.565s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.905s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14568599 8.80881528 + layer.9.1 0.03329684 4.43486738 + layer.19.0 0.11848472 24.88980744 + layer.19.1 0.11973745 44.25133044 + layer.29.0 0.10886538 69.07419971 + layer.29.1 0.10946879 132.58459945 + layer.39.0 14.08931437 1682.85131622 + layer.39.1 9.95616799 1616.05931102 + ------------------------------------------------------------------------------------- + TOTAL 3.08512769 447.86928087 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8921408 +BPFP 0.7079 bits/point +EBPFP 0.7079 equivalent bits/point +MSE 447.869281 +---------------------- --------------------------------------------------------- +Time: 21.750s Load: 1.280s, Pack+Encode: 7.565s, Decode+Unpack: 12.905s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 447.8693 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00332-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst (35/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.135s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 578,952B, BPFP=0.3675 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 570,992B, BPFP=0.3624 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,305,028B, BPFP=0.8284 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,321,056B, BPFP=0.8385 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,511,640B, BPFP=0.9595 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,528,068B, BPFP=0.9699 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,045,856B, BPFP=0.6639 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,051,168B, BPFP=0.6672 +⌛️ [2/4] FRONTEND: Frontend time: 7.690s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.436s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14493624 20.66611147 + layer.9.1 0.14482686 16.78760892 + layer.19.0 0.11946148 16.06546047 + layer.19.1 0.12828579 108.78800374 + layer.29.0 0.10467725 78.53552567 + layer.29.1 0.10613328 79.21205720 + layer.39.0 22.00188902 1642.88007800 + layer.39.1 19.26198661 1674.37942801 + ------------------------------------------------------------------------------------- + TOTAL 5.25152457 454.66428419 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8912760 +BPFP 0.7072 bits/point +EBPFP 0.7072 equivalent bits/point +MSE 454.664284 +---------------------- --------------------------------------------------------- +Time: 21.262s Load: 1.135s, Pack+Encode: 7.690s, Decode+Unpack: 12.436s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 454.6643 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00333-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst (36/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.236s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 574,716B, BPFP=0.3648 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 572,572B, BPFP=0.3634 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,269,324B, BPFP=0.8057 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,331,284B, BPFP=0.8450 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,495,440B, BPFP=0.9492 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,473,400B, BPFP=0.9352 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,037,416B, BPFP=0.6585 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,015,060B, BPFP=0.6443 +⌛️ [2/4] FRONTEND: Frontend time: 7.678s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.121s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14475438 16.81184494 + layer.9.1 0.14492096 17.27736406 + layer.19.0 0.11744098 20.62813439 + layer.19.1 0.11578254 6.60652307 + layer.29.0 0.11402616 138.39558620 + layer.29.1 0.11062706 98.92587138 + layer.39.0 28.92800668 1736.07653559 + layer.39.1 10.80449708 1589.37894053 + ------------------------------------------------------------------------------------- + TOTAL 5.06000698 453.01260002 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8769212 +BPFP 0.6958 bits/point +EBPFP 0.6958 equivalent bits/point +MSE 453.012600 +---------------------- --------------------------------------------------------- +Time: 22.035s Load: 1.236s, Pack+Encode: 7.678s, Decode+Unpack: 13.121s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 453.0126 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00334-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst (37/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.282s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 510,024B, BPFP=0.3237 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 525,792B, BPFP=0.3337 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,251,836B, BPFP=0.7946 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,281,160B, BPFP=0.8132 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,292,792B, BPFP=0.8206 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,296,568B, BPFP=0.8230 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 883,568B, BPFP=0.5608 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 892,412B, BPFP=0.5665 +⌛️ [2/4] FRONTEND: Frontend time: 7.671s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.812s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14546214 4.55714239 + layer.9.1 0.14553630 4.40453776 + layer.19.0 0.04765745 43.21684575 + layer.19.1 0.04191649 43.32919138 + layer.29.0 0.16505912 173.65835229 + layer.29.1 0.15755973 177.21508369 + layer.39.0 42.51041751 1542.84627884 + layer.39.1 31.38856333 1551.65388365 + ------------------------------------------------------------------------------------- + TOTAL 9.32527151 442.61016447 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7934152 +BPFP 0.6295 bits/point +EBPFP 0.6295 equivalent bits/point +MSE 442.610164 +---------------------- --------------------------------------------------------- +Time: 21.766s Load: 1.282s, Pack+Encode: 7.671s, Decode+Unpack: 12.812s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 442.6102 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01347-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst (38/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.279s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 536,324B, BPFP=0.3404 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 536,580B, BPFP=0.3406 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,251,016B, BPFP=0.7941 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,284,820B, BPFP=0.8155 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,424,016B, BPFP=0.9039 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,417,156B, BPFP=0.8995 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 942,476B, BPFP=0.5982 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 962,832B, BPFP=0.6112 +⌛️ [2/4] FRONTEND: Frontend time: 7.864s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.935s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03343166 4.58654242 + layer.9.1 0.03311388 4.43379592 + layer.19.0 0.03842411 6.25638685 + layer.19.1 0.03806642 15.78052258 + layer.29.0 4.26870163 72.23745227 + layer.29.1 4.26552788 37.12958543 + layer.39.0 33.95300821 1391.05784855 + layer.39.1 48.19954501 1430.28566786 + ------------------------------------------------------------------------------------- + TOTAL 11.35372735 370.22097523 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8355220 +BPFP 0.6629 bits/point +EBPFP 0.6629 equivalent bits/point +MSE 370.220975 +---------------------- --------------------------------------------------------- +Time: 22.078s Load: 1.279s, Pack+Encode: 7.864s, Decode+Unpack: 12.935s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 370.2210 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01390-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst (39/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.283s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 536,100B, BPFP=0.3403 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 540,604B, BPFP=0.3431 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,381,760B, BPFP=0.8771 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,325,672B, BPFP=0.8415 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,375,944B, BPFP=0.8734 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,388,392B, BPFP=0.8813 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 948,976B, BPFP=0.6024 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 968,588B, BPFP=0.6148 +⌛️ [2/4] FRONTEND: Frontend time: 7.640s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.680s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14549638 4.43715660 + layer.9.1 0.14520178 16.63868647 + layer.19.0 0.11487435 10.84851570 + layer.19.1 0.11481158 20.66723752 + layer.29.0 0.10827909 58.46349935 + layer.29.1 0.10618535 43.76834681 + layer.39.0 9.83978281 1505.39990250 + layer.39.1 9.67554703 1559.91160221 + ------------------------------------------------------------------------------------- + TOTAL 2.53127230 402.51686840 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8466036 +BPFP 0.6717 bits/point +EBPFP 0.6717 equivalent bits/point +MSE 402.516868 +---------------------- --------------------------------------------------------- +Time: 21.603s Load: 1.283s, Pack+Encode: 7.640s, Decode+Unpack: 12.680s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 402.5169 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01413-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst (40/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.228s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 525,804B, BPFP=0.3338 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 519,444B, BPFP=0.3297 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,311,044B, BPFP=0.8322 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,291,884B, BPFP=0.8200 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,446,696B, BPFP=0.9183 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,436,584B, BPFP=0.9119 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 939,264B, BPFP=0.5962 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 907,880B, BPFP=0.5763 +⌛️ [2/4] FRONTEND: Frontend time: 7.790s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.957s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00092318 4.45750953 + layer.9.1 0.00095285 4.44199562 + layer.19.0 0.08568402 10.92618241 + layer.19.1 0.08404610 19.99802085 + layer.29.0 0.12100375 92.84616713 + layer.29.1 0.12795564 78.90291883 + layer.39.0 12.85620633 1662.13649659 + layer.39.1 12.98640239 1682.59506012 + ------------------------------------------------------------------------------------- + TOTAL 3.28289678 444.53804389 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8378600 +BPFP 0.6648 bits/point +EBPFP 0.6648 equivalent bits/point +MSE 444.538044 +---------------------- --------------------------------------------------------- +Time: 21.975s Load: 1.228s, Pack+Encode: 7.790s, Decode+Unpack: 12.957s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 444.5380 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01421-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst (41/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.214s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 543,616B, BPFP=0.3451 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 533,116B, BPFP=0.3384 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,279,740B, BPFP=0.8123 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,309,764B, BPFP=0.8314 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,477,936B, BPFP=0.9381 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,464,224B, BPFP=0.9294 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,061,156B, BPFP=0.6736 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,062,444B, BPFP=0.6744 +⌛️ [2/4] FRONTEND: Frontend time: 7.882s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.217s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00113376 4.48073185 + layer.9.1 0.00100095 4.47461985 + layer.19.0 0.00983371 24.74915705 + layer.19.1 0.00806405 10.96449972 + layer.29.0 4.28365570 38.53127793 + layer.29.1 4.28597952 33.38904066 + layer.39.0 8.41906814 1474.48147546 + layer.39.1 8.59662605 1483.23643159 + ------------------------------------------------------------------------------------- + TOTAL 3.20067024 384.28840426 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8731996 +BPFP 0.6928 bits/point +EBPFP 0.6928 equivalent bits/point +MSE 384.288404 +---------------------- --------------------------------------------------------- +Time: 22.313s Load: 1.214s, Pack+Encode: 7.882s, Decode+Unpack: 13.217s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 384.2884 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01422-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst (42/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.284s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 576,248B, BPFP=0.3658 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 570,388B, BPFP=0.3621 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,294,844B, BPFP=0.8219 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,288,192B, BPFP=0.8177 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,527,324B, BPFP=0.9695 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,530,804B, BPFP=0.9717 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,084,696B, BPFP=0.6885 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,004,076B, BPFP=0.6373 +⌛️ [2/4] FRONTEND: Frontend time: 7.968s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.553s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14550294 4.43991903 + layer.9.1 0.14526658 8.63931742 + layer.19.0 0.11599200 10.96096035 + layer.19.1 0.11361485 20.37942167 + layer.29.0 4.26439454 29.25260501 + layer.29.1 4.25587461 29.25515925 + layer.39.0 8.37236706 1476.42557686 + layer.39.1 8.35116642 1466.58872278 + ------------------------------------------------------------------------------------- + TOTAL 3.22052237 380.74271030 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8876572 +BPFP 0.7043 bits/point +EBPFP 0.7043 equivalent bits/point +MSE 380.742710 +---------------------- --------------------------------------------------------- +Time: 21.804s Load: 1.284s, Pack+Encode: 7.968s, Decode+Unpack: 12.553s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 380.7427 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01441-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst (43/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.292s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 547,348B, BPFP=0.3474 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 566,840B, BPFP=0.3598 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,299,872B, BPFP=0.8251 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,295,836B, BPFP=0.8225 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,523,472B, BPFP=0.9670 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,533,660B, BPFP=0.9735 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,097,748B, BPFP=0.6968 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,096,488B, BPFP=0.6960 +⌛️ [2/4] FRONTEND: Frontend time: 7.781s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.629s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00080404 4.47968769 + layer.9.1 0.00082438 4.45894566 + layer.19.0 0.00843097 10.88385603 + layer.19.1 0.00674472 38.78946620 + layer.29.0 4.27713270 39.30970710 + layer.29.1 4.27133426 48.05113036 + layer.39.0 22.97048921 1492.86366591 + layer.39.1 18.06488920 1509.65030874 + ------------------------------------------------------------------------------------- + TOTAL 6.20008118 393.56084596 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8961264 +BPFP 0.7110 bits/point +EBPFP 0.7110 equivalent bits/point +MSE 393.560846 +---------------------- --------------------------------------------------------- +Time: 21.703s Load: 1.292s, Pack+Encode: 7.781s, Decode+Unpack: 12.629s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 393.5608 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00350-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst (44/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.278s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 555,080B, BPFP=0.3523 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 546,432B, BPFP=0.3468 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,293,204B, BPFP=0.8209 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,314,324B, BPFP=0.8343 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,518,728B, BPFP=0.9640 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,518,920B, BPFP=0.9641 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,100,680B, BPFP=0.6987 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,118,568B, BPFP=0.7100 +⌛️ [2/4] FRONTEND: Frontend time: 7.894s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.198s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14506060 12.66552750 + layer.9.1 0.14523201 4.39371115 + layer.19.0 0.04621643 29.70123091 + layer.19.1 0.04629335 29.41243500 + layer.29.0 4.27940669 64.19012025 + layer.29.1 4.27759670 58.91814775 + layer.39.0 19.91382637 1580.62463438 + layer.39.1 24.01088215 1571.43386415 + ------------------------------------------------------------------------------------- + TOTAL 6.60806429 418.91745889 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8965936 +BPFP 0.7114 bits/point +EBPFP 0.7114 equivalent bits/point +MSE 418.917459 +---------------------- --------------------------------------------------------- +Time: 22.370s Load: 1.278s, Pack+Encode: 7.894s, Decode+Unpack: 13.198s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 418.9175 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00351-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst (45/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.268s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 488,108B, BPFP=0.3098 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 498,440B, BPFP=0.3164 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,300,364B, BPFP=0.8254 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,262,028B, BPFP=0.8011 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,183,336B, BPFP=0.7511 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,158,400B, BPFP=0.7353 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 850,784B, BPFP=0.5400 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 857,448B, BPFP=0.5443 +⌛️ [2/4] FRONTEND: Frontend time: 7.813s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.453s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66985745 4.49780059 + layer.9.1 2.66884121 4.50259041 + layer.19.0 3.21935619 6.24124490 + layer.19.1 3.21606501 15.47766316 + layer.29.0 4.24164606 46.99195137 + layer.29.1 4.23648681 43.03484522 + layer.39.0 8.06392628 1261.24488138 + layer.39.1 8.17747540 1215.99642509 + ------------------------------------------------------------------------------------- + TOTAL 4.56170680 324.74842527 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7598908 +BPFP 0.6029 bits/point +EBPFP 0.6029 equivalent bits/point +MSE 324.748425 +---------------------- --------------------------------------------------------- +Time: 21.533s Load: 1.268s, Pack+Encode: 7.813s, Decode+Unpack: 12.453s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 324.7484 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00358-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst (46/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.251s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 489,648B, BPFP=0.3108 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 501,104B, BPFP=0.3181 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,224,584B, BPFP=0.7773 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,219,396B, BPFP=0.7740 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,276,484B, BPFP=0.8102 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,265,704B, BPFP=0.8034 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 937,620B, BPFP=0.5952 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 955,744B, BPFP=0.6067 +⌛️ [2/4] FRONTEND: Frontend time: 7.712s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.830s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66795352 4.52484315 + layer.9.1 2.66862889 4.51466748 + layer.19.0 3.22250645 6.43499274 + layer.19.1 3.22577319 11.11698311 + layer.29.0 4.25792136 38.21346888 + layer.29.1 4.25014663 49.32194508 + layer.39.0 8.65209937 1444.71628209 + layer.39.1 8.58450170 1436.34611635 + ------------------------------------------------------------------------------------- + TOTAL 4.69119139 374.39866236 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7870284 +BPFP 0.6245 bits/point +EBPFP 0.6245 equivalent bits/point +MSE 374.398662 +---------------------- --------------------------------------------------------- +Time: 21.793s Load: 1.251s, Pack+Encode: 7.712s, Decode+Unpack: 12.830s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 374.3987 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00360-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst (47/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.322s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 533,236B, BPFP=0.3385 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 533,144B, BPFP=0.3384 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,290,652B, BPFP=0.8192 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,284,444B, BPFP=0.8153 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,551,164B, BPFP=0.9846 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,548,536B, BPFP=0.9829 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,006,592B, BPFP=0.6389 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 983,148B, BPFP=0.6241 +⌛️ [2/4] FRONTEND: Frontend time: 7.698s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.070s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00092775 4.49883587 + layer.9.1 0.00093166 4.64385302 + layer.19.0 0.08227225 16.30288862 + layer.19.1 0.08381199 11.14539070 + layer.29.0 0.10725604 28.13066958 + layer.29.1 0.10756977 43.01563516 + layer.39.0 7.96294394 1303.12049074 + layer.39.1 7.95922050 1343.47221319 + ------------------------------------------------------------------------------------- + TOTAL 2.03811674 344.29124711 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8730916 +BPFP 0.6927 bits/point +EBPFP 0.6927 equivalent bits/point +MSE 344.291247 +---------------------- --------------------------------------------------------- +Time: 22.091s Load: 1.322s, Pack+Encode: 7.698s, Decode+Unpack: 13.070s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 344.2912 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00361-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst (48/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.278s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 484,172B, BPFP=0.3073 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 489,504B, BPFP=0.3107 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,380,484B, BPFP=0.8763 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,369,640B, BPFP=0.8694 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,260,108B, BPFP=0.7999 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,237,216B, BPFP=0.7853 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 877,724B, BPFP=0.5571 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 881,848B, BPFP=0.5598 +⌛️ [2/4] FRONTEND: Frontend time: 7.745s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.087s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66308118 4.53300350 + layer.9.1 2.66351027 4.54056464 + layer.19.0 3.21594155 6.54282085 + layer.19.1 3.21498593 6.42989252 + layer.29.0 4.33566519 198.45760887 + layer.29.1 4.34101296 188.67257069 + layer.39.0 8.65310735 1350.92378941 + layer.39.1 8.66575030 1370.54338642 + ------------------------------------------------------------------------------------- + TOTAL 4.71913184 391.33045461 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7980696 +BPFP 0.6332 bits/point +EBPFP 0.6332 equivalent bits/point +MSE 391.330455 +---------------------- --------------------------------------------------------- +Time: 22.110s Load: 1.278s, Pack+Encode: 7.745s, Decode+Unpack: 13.087s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 391.3305 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00384-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst (49/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.263s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 500,064B, BPFP=0.3174 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 498,160B, BPFP=0.3162 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,361,472B, BPFP=0.8642 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,360,152B, BPFP=0.8634 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,237,152B, BPFP=0.7853 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,234,252B, BPFP=0.7834 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 891,596B, BPFP=0.5659 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 897,364B, BPFP=0.5696 +⌛️ [2/4] FRONTEND: Frontend time: 8.045s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.482s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.65959259 4.53397117 + layer.9.1 2.65993726 4.53187015 + layer.19.0 3.20866700 29.60798109 + layer.19.1 3.21007805 6.44742495 + layer.29.0 4.27255361 79.67719166 + layer.29.1 4.27602442 129.05212057 + layer.39.0 19.11658068 1356.74796880 + layer.39.1 9.60360322 1443.71026974 + ------------------------------------------------------------------------------------- + TOTAL 6.12587960 381.78859977 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7980212 +BPFP 0.6332 bits/point +EBPFP 0.6332 equivalent bits/point +MSE 381.788600 +---------------------- --------------------------------------------------------- +Time: 21.791s Load: 1.263s, Pack+Encode: 8.045s, Decode+Unpack: 12.482s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 381.7886 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00385-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst (50/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.255s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 501,072B, BPFP=0.3181 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 492,904B, BPFP=0.3129 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,307,004B, BPFP=0.8296 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,248,472B, BPFP=0.7925 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,320,708B, BPFP=0.8383 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,312,632B, BPFP=0.8332 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 881,736B, BPFP=0.5597 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 875,024B, BPFP=0.5554 +⌛️ [2/4] FRONTEND: Frontend time: 7.904s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.147s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67078767 4.47707474 + layer.9.1 2.67131261 4.46503925 + layer.19.0 3.30595795 6.53180858 + layer.19.1 3.30543206 6.25696067 + layer.29.0 0.11228124 90.50010156 + layer.29.1 0.11507649 76.24158068 + layer.39.0 11.41791162 1446.87309067 + layer.39.1 11.38150745 1486.14234644 + ------------------------------------------------------------------------------------- + TOTAL 4.37253339 390.18600032 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7939552 +BPFP 0.6300 bits/point +EBPFP 0.6300 equivalent bits/point +MSE 390.186000 +---------------------- --------------------------------------------------------- +Time: 22.307s Load: 1.255s, Pack+Encode: 7.904s, Decode+Unpack: 13.147s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 390.1860 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00557-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst (51/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.256s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 532,432B, BPFP=0.3380 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 544,212B, BPFP=0.3454 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,303,352B, BPFP=0.8273 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,329,024B, BPFP=0.8436 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,408,540B, BPFP=0.8941 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,402,552B, BPFP=0.8903 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 916,404B, BPFP=0.5817 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 945,472B, BPFP=0.6001 +⌛️ [2/4] FRONTEND: Frontend time: 7.954s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.004s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14463971 4.37342868 + layer.9.1 0.14470460 4.51611407 + layer.19.0 0.12255537 15.69212352 + layer.19.1 0.11825690 10.70856607 + layer.29.0 0.11949990 90.76654412 + layer.29.1 0.11467140 104.34059149 + layer.39.0 10.68243977 1459.13227169 + layer.39.1 10.40156301 1488.47936302 + ------------------------------------------------------------------------------------- + TOTAL 2.73104133 397.25112533 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8381988 +BPFP 0.6651 bits/point +EBPFP 0.6651 equivalent bits/point +MSE 397.251125 +---------------------- --------------------------------------------------------- +Time: 22.214s Load: 1.256s, Pack+Encode: 7.954s, Decode+Unpack: 13.004s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 397.2511 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00001-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst (52/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.273s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 555,784B, BPFP=0.3528 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 553,880B, BPFP=0.3516 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,296,084B, BPFP=0.8227 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,273,072B, BPFP=0.8081 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,440,452B, BPFP=0.9143 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,436,996B, BPFP=0.9121 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 993,476B, BPFP=0.6306 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 955,368B, BPFP=0.6064 +⌛️ [2/4] FRONTEND: Frontend time: 7.824s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.811s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14485273 8.67268875 + layer.9.1 0.14484227 12.64480292 + layer.19.0 0.11969613 34.91581187 + layer.19.1 0.11916645 104.18402665 + layer.29.0 0.11480527 87.60667858 + layer.29.1 0.11451660 107.21074708 + layer.39.0 11.00270276 1529.47400065 + layer.39.1 11.01557422 1527.94670133 + ------------------------------------------------------------------------------------- + TOTAL 2.84701956 426.58193223 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8505112 +BPFP 0.6748 bits/point +EBPFP 0.6748 equivalent bits/point +MSE 426.581932 +---------------------- --------------------------------------------------------- +Time: 21.907s Load: 1.273s, Pack+Encode: 7.824s, Decode+Unpack: 12.811s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 426.5819 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00126-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst (53/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.255s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 533,844B, BPFP=0.3389 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 547,444B, BPFP=0.3475 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,261,628B, BPFP=0.8008 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,217,692B, BPFP=0.7729 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,296,180B, BPFP=0.8227 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,297,772B, BPFP=0.8238 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,007,236B, BPFP=0.6393 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 994,268B, BPFP=0.6311 +⌛️ [2/4] FRONTEND: Frontend time: 7.748s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.728s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14481488 4.39986093 + layer.9.1 0.14470567 4.38901750 + layer.19.0 0.03819180 39.83375650 + layer.19.1 0.04002141 16.19598280 + layer.29.0 0.11241068 105.78357572 + layer.29.1 0.11133552 109.87710229 + layer.39.0 31.78807483 1757.97595060 + layer.39.1 43.50691623 1688.27364316 + ------------------------------------------------------------------------------------- + TOTAL 9.48580888 465.84111119 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8156064 +BPFP 0.6471 bits/point +EBPFP 0.6471 equivalent bits/point +MSE 465.841111 +---------------------- --------------------------------------------------------- +Time: 21.732s Load: 1.255s, Pack+Encode: 7.748s, Decode+Unpack: 12.728s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 465.8411 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00127-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst (54/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.237s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 540,156B, BPFP=0.3429 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 546,924B, BPFP=0.3472 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,324,492B, BPFP=0.8407 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,299,040B, BPFP=0.8246 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,389,108B, BPFP=0.8817 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,407,084B, BPFP=0.8931 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 953,168B, BPFP=0.6050 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 942,536B, BPFP=0.5983 +⌛️ [2/4] FRONTEND: Frontend time: 7.752s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.478s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14528272 4.40328984 + layer.9.1 0.14516892 4.40544038 + layer.19.0 0.11319376 67.23622339 + layer.19.1 0.11666145 39.24499055 + layer.29.0 0.21118872 205.19318736 + layer.29.1 0.20646930 228.02165258 + layer.39.0 14.37750853 1873.59213520 + layer.39.1 21.76644002 1797.09473513 + ------------------------------------------------------------------------------------- + TOTAL 4.63523918 527.39895680 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8402508 +BPFP 0.6667 bits/point +EBPFP 0.6667 equivalent bits/point +MSE 527.398957 +---------------------- --------------------------------------------------------- +Time: 21.467s Load: 1.237s, Pack+Encode: 7.752s, Decode+Unpack: 12.478s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 527.3990 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00128-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst (55/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.208s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 522,840B, BPFP=0.3319 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 511,392B, BPFP=0.3246 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,313,000B, BPFP=0.8334 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,282,728B, BPFP=0.8142 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,350,376B, BPFP=0.8572 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,338,724B, BPFP=0.8498 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 943,452B, BPFP=0.5989 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 918,684B, BPFP=0.5831 +⌛️ [2/4] FRONTEND: Frontend time: 7.812s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.938s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14454765 4.39943120 + layer.9.1 0.14475082 4.40683270 + layer.19.0 0.04087094 34.40326414 + layer.19.1 0.11687931 29.46336478 + layer.29.0 0.10817139 53.58186241 + layer.29.1 0.10802081 122.00859197 + layer.39.0 19.80422286 1577.95336367 + layer.39.1 34.29222355 1512.08547286 + ------------------------------------------------------------------------------------- + TOTAL 6.84496092 417.28777297 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8181196 +BPFP 0.6491 bits/point +EBPFP 0.6491 equivalent bits/point +MSE 417.287773 +---------------------- --------------------------------------------------------- +Time: 21.957s Load: 1.208s, Pack+Encode: 7.812s, Decode+Unpack: 12.938s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 417.2878 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00131-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst (56/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.265s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 497,984B, BPFP=0.3161 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 499,544B, BPFP=0.3171 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,300,852B, BPFP=0.8257 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,275,832B, BPFP=0.8098 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,265,632B, BPFP=0.8034 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,277,088B, BPFP=0.8106 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 882,196B, BPFP=0.5600 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 900,552B, BPFP=0.5716 +⌛️ [2/4] FRONTEND: Frontend time: 8.846s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.117s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14462205 4.41659991 + layer.9.1 0.14495783 4.42825582 + layer.19.0 0.04322015 15.92418295 + layer.19.1 0.03788725 52.84439998 + layer.29.0 0.10021623 53.67826820 + layer.29.1 0.10137775 63.68805858 + layer.39.0 58.66958482 1405.48781280 + layer.39.1 72.48303949 1415.42200195 + ------------------------------------------------------------------------------------- + TOTAL 16.46561320 376.98619753 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7899680 +BPFP 0.6268 bits/point +EBPFP 0.6268 equivalent bits/point +MSE 376.986198 +---------------------- --------------------------------------------------------- +Time: 23.228s Load: 1.265s, Pack+Encode: 8.846s, Decode+Unpack: 13.117s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 376.9862 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00132-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst (57/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.282s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 544,672B, BPFP=0.3457 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 550,032B, BPFP=0.3491 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,270,948B, BPFP=0.8067 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,284,708B, BPFP=0.8155 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,456,628B, BPFP=0.9246 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,476,308B, BPFP=0.9371 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 922,564B, BPFP=0.5856 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 938,892B, BPFP=0.5960 +⌛️ [2/4] FRONTEND: Frontend time: 7.749s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.106s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14527927 4.39577599 + layer.9.1 0.14528875 4.54456451 + layer.19.0 0.12591341 43.18038065 + layer.19.1 0.13556211 20.54239874 + layer.29.0 0.11238900 71.70316055 + layer.29.1 0.11028371 111.77978754 + layer.39.0 11.48751193 1441.11325967 + layer.39.1 11.29491489 1513.36935327 + ------------------------------------------------------------------------------------- + TOTAL 2.94464289 401.32858511 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8444752 +BPFP 0.6700 bits/point +EBPFP 0.6700 equivalent bits/point +MSE 401.328585 +---------------------- --------------------------------------------------------- +Time: 22.136s Load: 1.282s, Pack+Encode: 7.749s, Decode+Unpack: 13.106s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 401.3286 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00136-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst (58/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.283s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 549,620B, BPFP=0.3489 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 554,724B, BPFP=0.3521 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,287,596B, BPFP=0.8173 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,232,756B, BPFP=0.7825 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,412,420B, BPFP=0.8965 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,414,024B, BPFP=0.8976 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 929,080B, BPFP=0.5897 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 911,120B, BPFP=0.5783 +⌛️ [2/4] FRONTEND: Frontend time: 7.880s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.185s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14489184 4.41856097 + layer.9.1 0.14511764 4.40364435 + layer.19.0 0.03976490 24.81246699 + layer.19.1 0.11370806 10.72514574 + layer.29.0 0.10933599 56.28114336 + layer.29.1 0.11012027 90.35920743 + layer.39.0 9.10787636 1243.99658759 + layer.39.1 9.00026152 1329.29484888 + ------------------------------------------------------------------------------------- + TOTAL 2.34638457 345.53645066 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8291340 +BPFP 0.6579 bits/point +EBPFP 0.6579 equivalent bits/point +MSE 345.536451 +---------------------- --------------------------------------------------------- +Time: 22.348s Load: 1.283s, Pack+Encode: 7.880s, Decode+Unpack: 13.185s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 345.5365 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00193-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst (59/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.281s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 498,184B, BPFP=0.3162 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 488,260B, BPFP=0.3099 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,266,572B, BPFP=0.8040 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,245,212B, BPFP=0.7904 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,228,996B, BPFP=0.7801 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,225,760B, BPFP=0.7781 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 875,152B, BPFP=0.5555 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 853,328B, BPFP=0.5416 +⌛️ [2/4] FRONTEND: Frontend time: 7.806s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.696s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00083307 4.44190961 + layer.9.1 0.00247171 4.43243374 + layer.19.0 0.00642632 6.18773994 + layer.19.1 0.00641681 6.48223653 + layer.29.0 0.10256791 42.71623131 + layer.29.1 0.10162673 32.76376645 + layer.39.0 8.50517638 1126.62455314 + layer.39.1 8.55767781 1164.18963276 + ------------------------------------------------------------------------------------- + TOTAL 2.16039959 298.47981293 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7681464 +BPFP 0.6095 bits/point +EBPFP 0.6095 equivalent bits/point +MSE 298.479813 +---------------------- --------------------------------------------------------- +Time: 21.783s Load: 1.281s, Pack+Encode: 7.806s, Decode+Unpack: 12.696s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 298.4798 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00196-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst (60/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.238s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 522,320B, BPFP=0.3315 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 521,544B, BPFP=0.3310 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,329,056B, BPFP=0.8436 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,315,396B, BPFP=0.8349 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,411,544B, BPFP=0.8960 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,404,056B, BPFP=0.8912 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 977,472B, BPFP=0.6205 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 949,556B, BPFP=0.6027 +⌛️ [2/4] FRONTEND: Frontend time: 7.932s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.530s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00064391 4.45653519 + layer.9.1 0.00065402 4.45011660 + layer.19.0 0.08134466 20.17351900 + layer.19.1 0.08141702 6.05160896 + layer.29.0 0.11551180 140.66841688 + layer.29.1 0.11251285 138.77716729 + layer.39.0 10.61319619 1434.49155021 + layer.39.1 10.43102047 1478.59831004 + ------------------------------------------------------------------------------------- + TOTAL 2.67953762 403.45840302 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8430944 +BPFP 0.6689 bits/point +EBPFP 0.6689 equivalent bits/point +MSE 403.458403 +---------------------- --------------------------------------------------------- +Time: 21.700s Load: 1.238s, Pack+Encode: 7.932s, Decode+Unpack: 12.530s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 403.4584 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00559-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst (61/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.258s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 539,232B, BPFP=0.3423 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 551,968B, BPFP=0.3504 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,316,560B, BPFP=0.8357 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,313,352B, BPFP=0.8336 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,461,468B, BPFP=0.9277 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,472,052B, BPFP=0.9344 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 968,480B, BPFP=0.6147 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 975,708B, BPFP=0.6193 +⌛️ [2/4] FRONTEND: Frontend time: 7.814s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.168s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14441776 4.36717830 + layer.9.1 0.14449203 4.36823103 + layer.19.0 0.11315974 29.65906829 + layer.19.1 0.11435745 6.48676610 + layer.29.0 0.12811458 131.05382678 + layer.29.1 0.12952277 159.58263934 + layer.39.0 31.10682331 1665.01917452 + layer.39.1 16.99297713 1711.77543061 + ------------------------------------------------------------------------------------- + TOTAL 6.10923309 464.03903937 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8598820 +BPFP 0.6823 bits/point +EBPFP 0.6823 equivalent bits/point +MSE 464.039039 +---------------------- --------------------------------------------------------- +Time: 22.240s Load: 1.258s, Pack+Encode: 7.814s, Decode+Unpack: 13.168s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 464.0390 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00570-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst (62/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.283s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 524,572B, BPFP=0.3330 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 521,452B, BPFP=0.3310 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,314,652B, BPFP=0.8345 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,286,148B, BPFP=0.8164 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,391,172B, BPFP=0.8830 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,381,716B, BPFP=0.8770 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 896,128B, BPFP=0.5688 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 910,532B, BPFP=0.5780 +⌛️ [2/4] FRONTEND: Frontend time: 8.003s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.959s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00078516 4.47576430 + layer.9.1 0.00079184 4.48883063 + layer.19.0 3.22632161 6.20623337 + layer.19.1 3.22513146 6.13583772 + layer.29.0 0.10494786 61.69621384 + layer.29.1 0.10251782 51.40915461 + layer.39.0 10.88842496 1465.56581085 + layer.39.1 10.78217420 1394.90136497 + ------------------------------------------------------------------------------------- + TOTAL 3.54138686 374.35990129 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8226372 +BPFP 0.6527 bits/point +EBPFP 0.6527 equivalent bits/point +MSE 374.359901 +---------------------- --------------------------------------------------------- +Time: 22.245s Load: 1.283s, Pack+Encode: 8.003s, Decode+Unpack: 12.959s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 374.3599 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00758-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst (63/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.255s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 561,616B, BPFP=0.3565 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 531,208B, BPFP=0.3372 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,278,708B, BPFP=0.8117 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,266,372B, BPFP=0.8038 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,302,876B, BPFP=0.8270 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,294,956B, BPFP=0.8220 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 923,984B, BPFP=0.5865 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 932,816B, BPFP=0.5921 +⌛️ [2/4] FRONTEND: Frontend time: 7.807s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.776s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14561636 4.47828553 + layer.9.1 0.14552785 4.59436825 + layer.19.0 0.04069186 11.39499462 + layer.19.1 0.03840616 6.33114006 + layer.29.0 0.11346353 55.37978347 + layer.29.1 0.11182956 41.49609248 + layer.39.0 10.19697364 1438.80646734 + layer.39.1 10.11578978 1507.92102697 + ------------------------------------------------------------------------------------- + TOTAL 2.61353734 383.80026984 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8092536 +BPFP 0.6421 bits/point +EBPFP 0.6421 equivalent bits/point +MSE 383.800270 +---------------------- --------------------------------------------------------- +Time: 21.838s Load: 1.255s, Pack+Encode: 7.807s, Decode+Unpack: 12.776s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 383.8003 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00764-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst (64/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.257s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 545,528B, BPFP=0.3463 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 531,264B, BPFP=0.3372 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,287,148B, BPFP=0.8170 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,282,620B, BPFP=0.8141 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,431,724B, BPFP=0.9088 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,422,588B, BPFP=0.9030 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 932,776B, BPFP=0.5921 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 936,484B, BPFP=0.5944 +⌛️ [2/4] FRONTEND: Frontend time: 7.731s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.323s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14399445 4.58836097 + layer.9.1 0.14558028 4.43138990 + layer.19.0 0.03837104 6.74517781 + layer.19.1 0.04376782 11.04516498 + layer.29.0 0.11695251 106.34903112 + layer.29.1 0.13128335 133.12984441 + layer.39.0 11.28613757 1448.11797205 + layer.39.1 11.84408769 1407.31589210 + ------------------------------------------------------------------------------------- + TOTAL 2.96877184 390.21535417 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8370132 +BPFP 0.6641 bits/point +EBPFP 0.6641 equivalent bits/point +MSE 390.215354 +---------------------- --------------------------------------------------------- +Time: 21.311s Load: 1.257s, Pack+Encode: 7.731s, Decode+Unpack: 12.323s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 390.2154 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00845-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst (65/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.226s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 550,456B, BPFP=0.3494 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 525,364B, BPFP=0.3335 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,279,504B, BPFP=0.8122 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,330,656B, BPFP=0.8446 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,560,152B, BPFP=0.9903 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,550,768B, BPFP=0.9843 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 962,504B, BPFP=0.6109 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 982,376B, BPFP=0.6236 +⌛️ [2/4] FRONTEND: Frontend time: 7.728s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.840s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14502686 4.38443111 + layer.9.1 0.03259508 4.39894847 + layer.19.0 0.11326540 20.30706654 + layer.19.1 0.11324834 29.72786755 + layer.29.0 0.12250664 117.21029006 + layer.29.1 0.12058897 145.93266575 + layer.39.0 16.17915050 1791.83230419 + layer.39.1 21.66230805 1814.56321092 + ------------------------------------------------------------------------------------- + TOTAL 4.81108623 491.04459807 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8741780 +BPFP 0.6936 bits/point +EBPFP 0.6936 equivalent bits/point +MSE 491.044598 +---------------------- --------------------------------------------------------- +Time: 21.795s Load: 1.226s, Pack+Encode: 7.728s, Decode+Unpack: 12.840s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 491.0446 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00870-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst (66/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.257s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 510,348B, BPFP=0.3239 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 510,100B, BPFP=0.3238 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,310,208B, BPFP=0.8317 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,290,640B, BPFP=0.8192 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,338,324B, BPFP=0.8495 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,338,348B, BPFP=0.8495 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 958,408B, BPFP=0.6083 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 930,476B, BPFP=0.5906 +⌛️ [2/4] FRONTEND: Frontend time: 8.106s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.812s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66733019 4.52263359 + layer.9.1 2.66763138 4.53540253 + layer.19.0 3.22293078 6.62306719 + layer.19.1 3.22376992 6.41009278 + layer.29.0 4.27658332 30.84849285 + layer.29.1 4.27160529 30.62232643 + layer.39.0 7.81683598 1438.06402340 + layer.39.1 9.86231960 1324.69564511 + ------------------------------------------------------------------------------------- + TOTAL 4.75112581 355.79021048 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8186852 +BPFP 0.6496 bits/point +EBPFP 0.6496 equivalent bits/point +MSE 355.790210 +---------------------- --------------------------------------------------------- +Time: 22.174s Load: 1.257s, Pack+Encode: 8.106s, Decode+Unpack: 12.812s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 355.7902 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00153-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst (67/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.245s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 493,144B, BPFP=0.3130 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 486,656B, BPFP=0.3089 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,225,164B, BPFP=0.7777 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,230,744B, BPFP=0.7812 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,234,104B, BPFP=0.7833 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,234,248B, BPFP=0.7834 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 847,068B, BPFP=0.5377 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 870,004B, BPFP=0.5522 +⌛️ [2/4] FRONTEND: Frontend time: 7.966s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.107s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14549763 4.42973098 + layer.9.1 0.14520254 4.41894436 + layer.19.0 0.04746155 11.42562764 + layer.19.1 0.04383140 30.13316542 + layer.29.0 4.26247378 59.47122299 + layer.29.1 4.25497898 36.06498060 + layer.39.0 7.94138086 1343.23927527 + layer.39.1 7.86439079 1314.80736107 + ------------------------------------------------------------------------------------- + TOTAL 3.08815219 350.49878854 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7621132 +BPFP 0.6047 bits/point +EBPFP 0.6047 equivalent bits/point +MSE 350.498789 +---------------------- --------------------------------------------------------- +Time: 22.318s Load: 1.245s, Pack+Encode: 7.966s, Decode+Unpack: 13.107s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 350.4988 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00167-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst (68/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.281s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 487,540B, BPFP=0.3095 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 499,680B, BPFP=0.3172 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,205,572B, BPFP=0.7652 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,117,112B, BPFP=0.7091 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,200,232B, BPFP=0.7618 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,199,988B, BPFP=0.7617 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 889,264B, BPFP=0.5645 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 900,888B, BPFP=0.5718 +⌛️ [2/4] FRONTEND: Frontend time: 7.998s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.585s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11283708 8.56041548 + layer.9.1 0.11300174 4.46068614 + layer.19.0 3.22718329 11.39026954 + layer.19.1 3.22892155 6.51454973 + layer.29.0 4.26448309 58.52078425 + layer.29.1 4.25758082 39.85930137 + layer.39.0 9.82393946 1436.74634384 + layer.39.1 9.78394007 1384.34075398 + ------------------------------------------------------------------------------------- + TOTAL 4.35148589 368.79913804 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7500276 +BPFP 0.5951 bits/point +EBPFP 0.5951 equivalent bits/point +MSE 368.799138 +---------------------- --------------------------------------------------------- +Time: 21.864s Load: 1.281s, Pack+Encode: 7.998s, Decode+Unpack: 12.585s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 368.7991 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00208-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst (69/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.258s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 586,816B, BPFP=0.3725 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 570,604B, BPFP=0.3622 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,256,000B, BPFP=0.7972 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,285,872B, BPFP=0.8162 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,418,200B, BPFP=0.9002 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,419,452B, BPFP=0.9010 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 991,276B, BPFP=0.6292 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,015,360B, BPFP=0.6445 +⌛️ [2/4] FRONTEND: Frontend time: 7.844s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.901s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14503861 8.69117711 + layer.9.1 0.14483112 37.25671057 + layer.19.0 0.11529889 30.32981597 + layer.19.1 0.11517203 43.93355947 + layer.29.0 0.11961639 70.81063130 + layer.29.1 0.11795276 79.90241103 + layer.39.0 83.84633978 1945.92606435 + layer.39.1 174.87768118 1921.66818330 + ------------------------------------------------------------------------------------- + TOTAL 32.43524135 517.31481914 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8543580 +BPFP 0.6779 bits/point +EBPFP 0.6779 equivalent bits/point +MSE 517.314819 +---------------------- --------------------------------------------------------- +Time: 22.003s Load: 1.258s, Pack+Encode: 7.844s, Decode+Unpack: 12.901s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 517.3148 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01201-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst (70/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.260s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 519,076B, BPFP=0.3295 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 520,280B, BPFP=0.3302 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,344,900B, BPFP=0.8537 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,330,152B, BPFP=0.8443 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,324,172B, BPFP=0.8405 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,321,624B, BPFP=0.8389 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 951,496B, BPFP=0.6040 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 935,832B, BPFP=0.5940 +⌛️ [2/4] FRONTEND: Frontend time: 8.075s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.840s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.81453913 4.59742012 + layer.9.1 0.14528001 4.43667038 + layer.19.0 3.26598681 33.86273663 + layer.19.1 0.04116655 15.11309083 + layer.29.0 4.28557138 92.52062683 + layer.29.1 4.28198282 82.47422408 + layer.39.0 74.89367180 1526.53997400 + layer.39.1 42.04871577 1615.25316867 + ------------------------------------------------------------------------------------- + TOTAL 16.47211428 421.84973894 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8247532 +BPFP 0.6544 bits/point +EBPFP 0.6544 equivalent bits/point +MSE 421.849739 +---------------------- --------------------------------------------------------- +Time: 22.175s Load: 1.260s, Pack+Encode: 8.075s, Decode+Unpack: 12.840s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 421.8497 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01296-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst (71/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.257s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 530,764B, BPFP=0.3369 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 517,152B, BPFP=0.3283 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,233,604B, BPFP=0.7830 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,322,880B, BPFP=0.8397 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,320,036B, BPFP=0.8379 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,327,548B, BPFP=0.8427 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 867,748B, BPFP=0.5508 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 861,808B, BPFP=0.5470 +⌛️ [2/4] FRONTEND: Frontend time: 7.783s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.792s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66710422 4.48790008 + layer.9.1 2.66812426 4.47648855 + layer.19.0 3.22059776 11.04223498 + layer.19.1 3.22546153 6.48981861 + layer.29.0 0.11226317 70.76036927 + layer.29.1 0.11257672 42.10818675 + layer.39.0 59.39237691 1516.21644459 + layer.39.1 37.52358222 1498.38804030 + ------------------------------------------------------------------------------------- + TOTAL 13.61526085 394.24618539 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7981540 +BPFP 0.6333 bits/point +EBPFP 0.6333 equivalent bits/point +MSE 394.246185 +---------------------- --------------------------------------------------------- +Time: 21.833s Load: 1.257s, Pack+Encode: 7.783s, Decode+Unpack: 12.792s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 394.2462 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01313-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst (72/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.245s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 516,892B, BPFP=0.3281 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 506,732B, BPFP=0.3216 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,325,340B, BPFP=0.8413 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,348,768B, BPFP=0.8561 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,380,988B, BPFP=0.8766 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,392,772B, BPFP=0.8841 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 854,068B, BPFP=0.5421 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 854,436B, BPFP=0.5424 +⌛️ [2/4] FRONTEND: Frontend time: 8.074s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.731s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14519123 4.41351312 + layer.9.1 0.14511500 4.39956513 + layer.19.0 0.03974548 6.50007871 + layer.19.1 0.03981401 11.49725407 + layer.29.0 4.26343511 42.20590520 + layer.29.1 4.25610090 75.99108304 + layer.39.0 7.90972018 1269.10350991 + layer.39.1 8.05601540 1245.82580435 + ------------------------------------------------------------------------------------- + TOTAL 3.10689216 332.49208919 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8179996 +BPFP 0.6490 bits/point +EBPFP 0.6490 equivalent bits/point +MSE 332.492089 +---------------------- --------------------------------------------------------- +Time: 22.050s Load: 1.245s, Pack+Encode: 8.074s, Decode+Unpack: 12.731s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 332.4921 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00032-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst (73/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.244s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 508,656B, BPFP=0.3229 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 510,980B, BPFP=0.3243 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,210,568B, BPFP=0.7684 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,189,964B, BPFP=0.7553 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,428,504B, BPFP=0.9067 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,413,064B, BPFP=0.8969 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 900,492B, BPFP=0.5716 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 868,108B, BPFP=0.5510 +⌛️ [2/4] FRONTEND: Frontend time: 7.629s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.995s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14586543 4.54205915 + layer.9.1 0.14572574 4.41713405 + layer.19.0 0.03953905 20.32645053 + layer.19.1 0.03760033 15.73198072 + layer.29.0 0.10448607 45.54729647 + layer.29.1 0.10697372 66.14855683 + layer.39.0 14.19073468 1439.87357816 + layer.39.1 8.92149669 1522.46295093 + ------------------------------------------------------------------------------------- + TOTAL 2.96155271 389.88125086 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8030336 +BPFP 0.6372 bits/point +EBPFP 0.6372 equivalent bits/point +MSE 389.881251 +---------------------- --------------------------------------------------------- +Time: 21.868s Load: 1.244s, Pack+Encode: 7.629s, Decode+Unpack: 12.995s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 389.8813 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00041-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst (74/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.235s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 530,808B, BPFP=0.3369 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 527,048B, BPFP=0.3345 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,280,836B, BPFP=0.8130 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,315,168B, BPFP=0.8348 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,434,416B, BPFP=0.9105 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,424,120B, BPFP=0.9040 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 920,212B, BPFP=0.5841 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 904,364B, BPFP=0.5740 +⌛️ [2/4] FRONTEND: Frontend time: 7.620s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.906s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00251883 4.45555768 + layer.9.1 0.14409062 4.45309358 + layer.19.0 0.12740102 89.74230175 + layer.19.1 0.12254588 38.89445889 + layer.29.0 4.25147928 32.16489529 + layer.29.1 4.25065697 52.26903741 + layer.39.0 9.21805114 1398.73740656 + layer.39.1 9.03214690 1476.42768931 + ------------------------------------------------------------------------------------- + TOTAL 3.39361133 387.14305506 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8336972 +BPFP 0.6615 bits/point +EBPFP 0.6615 equivalent bits/point +MSE 387.143055 +---------------------- --------------------------------------------------------- +Time: 21.761s Load: 1.235s, Pack+Encode: 7.620s, Decode+Unpack: 12.906s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 387.1431 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00042-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst (75/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.227s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 564,776B, BPFP=0.3585 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 563,484B, BPFP=0.3577 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,367,964B, BPFP=0.8683 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,304,660B, BPFP=0.8281 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,511,088B, BPFP=0.9592 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,509,232B, BPFP=0.9580 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,092,976B, BPFP=0.6938 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,045,716B, BPFP=0.6638 +⌛️ [2/4] FRONTEND: Frontend time: 7.683s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.701s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14550905 12.62697153 + layer.9.1 0.14590163 12.85298561 + layer.19.0 0.12839093 39.52540776 + layer.19.1 0.12422524 44.02909693 + layer.29.0 0.11695262 98.79681914 + layer.29.1 0.11389293 117.84642103 + layer.39.0 10.18180439 1613.81329217 + layer.39.1 10.42432323 1674.92671433 + ------------------------------------------------------------------------------------- + TOTAL 2.67262500 451.80221356 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8959896 +BPFP 0.7109 bits/point +EBPFP 0.7109 equivalent bits/point +MSE 451.802214 +---------------------- --------------------------------------------------------- +Time: 21.611s Load: 1.227s, Pack+Encode: 7.683s, Decode+Unpack: 12.701s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 451.8022 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00620-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst (76/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.281s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 533,536B, BPFP=0.3387 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 539,732B, BPFP=0.3426 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,286,592B, BPFP=0.8167 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,301,072B, BPFP=0.8259 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,480,296B, BPFP=0.9396 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,490,492B, BPFP=0.9461 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 907,052B, BPFP=0.5758 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 909,296B, BPFP=0.5772 +⌛️ [2/4] FRONTEND: Frontend time: 7.989s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.525s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14509775 4.41934584 + layer.9.1 0.14508723 12.48864433 + layer.19.0 0.11633494 20.38376589 + layer.19.1 0.11804005 15.47192756 + layer.29.0 0.15409572 77.43477819 + layer.29.1 0.14997486 114.78413430 + layer.39.0 9.23291952 1527.43711407 + layer.39.1 9.22304726 1583.60107247 + ------------------------------------------------------------------------------------- + TOTAL 2.41057467 419.50259783 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8448068 +BPFP 0.6703 bits/point +EBPFP 0.6703 equivalent bits/point +MSE 419.502598 +---------------------- --------------------------------------------------------- +Time: 21.794s Load: 1.281s, Pack+Encode: 7.989s, Decode+Unpack: 12.525s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 419.5026 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00636-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst (77/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.281s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 550,532B, BPFP=0.3495 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 556,792B, BPFP=0.3534 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,302,724B, BPFP=0.8269 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,290,764B, BPFP=0.8193 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,487,564B, BPFP=0.9442 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,500,596B, BPFP=0.9525 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 939,696B, BPFP=0.5965 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 943,288B, BPFP=0.5988 +⌛️ [2/4] FRONTEND: Frontend time: 7.947s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.933s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14497060 12.83904397 + layer.9.1 0.14492971 16.63628204 + layer.19.0 0.11929473 39.10739712 + layer.19.1 0.11869117 44.00364092 + layer.29.0 0.13715227 159.30709701 + layer.29.1 0.14278979 163.05654859 + layer.39.0 9.99110525 1685.66200845 + layer.39.1 10.01170034 1631.56158596 + ------------------------------------------------------------------------------------- + TOTAL 2.60132923 469.02170051 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8571956 +BPFP 0.6801 bits/point +EBPFP 0.6801 equivalent bits/point +MSE 469.021701 +---------------------- --------------------------------------------------------- +Time: 22.162s Load: 1.281s, Pack+Encode: 7.947s, Decode+Unpack: 12.933s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 469.0217 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00637-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst (78/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.284s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 546,572B, BPFP=0.3469 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 552,264B, BPFP=0.3505 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,314,308B, BPFP=0.8343 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,334,720B, BPFP=0.8472 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,478,124B, BPFP=0.9382 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,484,708B, BPFP=0.9424 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,058,788B, BPFP=0.6721 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,017,892B, BPFP=0.6461 +⌛️ [2/4] FRONTEND: Frontend time: 8.050s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.584s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14589856 12.52644748 + layer.9.1 0.03321603 4.45817380 + layer.19.0 0.11866178 20.47017692 + layer.19.1 0.11267978 38.69050871 + layer.29.0 0.10803594 58.16420214 + layer.29.1 0.10714094 78.17273318 + layer.39.0 11.58943751 1635.10172246 + layer.39.1 9.70079103 1644.59506012 + ------------------------------------------------------------------------------------- + TOTAL 2.73948269 436.52237810 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8787376 +BPFP 0.6972 bits/point +EBPFP 0.6972 equivalent bits/point +MSE 436.522378 +---------------------- --------------------------------------------------------- +Time: 21.917s Load: 1.284s, Pack+Encode: 8.050s, Decode+Unpack: 12.584s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 436.5224 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00410-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst (79/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.270s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 516,256B, BPFP=0.3277 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 524,892B, BPFP=0.3332 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,359,820B, BPFP=0.8631 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,368,200B, BPFP=0.8685 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,398,620B, BPFP=0.8878 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,398,940B, BPFP=0.8880 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 914,856B, BPFP=0.5807 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 928,580B, BPFP=0.5894 +⌛️ [2/4] FRONTEND: Frontend time: 7.883s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.149s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03308301 4.56145647 + layer.9.1 0.14566304 4.56263425 + layer.19.0 0.03810260 29.32966617 + layer.19.1 0.03780774 15.39032794 + layer.29.0 0.11592613 81.21203689 + layer.29.1 0.11717217 90.79588479 + layer.39.0 9.98032847 1419.13974651 + layer.39.1 9.70849498 1383.91663958 + ------------------------------------------------------------------------------------- + TOTAL 2.52207227 378.61354907 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8410164 +BPFP 0.6673 bits/point +EBPFP 0.6673 equivalent bits/point +MSE 378.613549 +---------------------- --------------------------------------------------------- +Time: 22.302s Load: 1.270s, Pack+Encode: 7.883s, Decode+Unpack: 13.149s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 378.6135 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00411-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst (80/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.276s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 504,384B, BPFP=0.3202 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 503,840B, BPFP=0.3198 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,191,736B, BPFP=0.7565 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,238,840B, BPFP=0.7864 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,284,300B, BPFP=0.8152 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,281,220B, BPFP=0.8133 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 894,436B, BPFP=0.5677 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 899,244B, BPFP=0.5708 +⌛️ [2/4] FRONTEND: Frontend time: 7.747s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.685s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14555504 4.43430086 + layer.9.1 0.14557384 8.63123134 + layer.19.0 0.03995539 29.34982329 + layer.19.1 0.04542811 29.40944660 + layer.29.0 0.12033866 65.36597132 + layer.29.1 0.13252172 75.74424663 + layer.39.0 10.37566776 1302.11683458 + layer.39.1 9.84188447 1275.88552161 + ------------------------------------------------------------------------------------- + TOTAL 2.60586562 348.86717203 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 7798000 +BPFP 0.6187 bits/point +EBPFP 0.6187 equivalent bits/point +MSE 348.867172 +---------------------- --------------------------------------------------------- +Time: 21.708s Load: 1.276s, Pack+Encode: 7.747s, Decode+Unpack: 12.685s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 348.8672 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00412-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst (81/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.254s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 557,836B, BPFP=0.3541 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 561,792B, BPFP=0.3566 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,298,860B, BPFP=0.8245 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,343,036B, BPFP=0.8525 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,526,248B, BPFP=0.9688 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,524,324B, BPFP=0.9676 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,153,644B, BPFP=0.7323 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,159,312B, BPFP=0.7359 +⌛️ [2/4] FRONTEND: Frontend time: 7.825s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.598s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14470203 8.53045085 + layer.9.1 0.14481130 4.52054653 + layer.19.0 0.11257574 43.90066420 + layer.19.1 0.11422884 38.70392225 + layer.29.0 0.10456927 63.38414040 + layer.29.1 0.10551051 53.97232491 + layer.39.0 10.36536069 1788.17370816 + layer.39.1 11.81531702 1705.53152421 + ------------------------------------------------------------------------------------- + TOTAL 2.86338443 463.33966019 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9125052 +BPFP 0.7240 bits/point +EBPFP 0.7240 equivalent bits/point +MSE 463.339660 +---------------------- --------------------------------------------------------- +Time: 21.678s Load: 1.254s, Pack+Encode: 7.825s, Decode+Unpack: 12.598s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 463.3397 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00413-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst (82/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.259s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 558,412B, BPFP=0.3545 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 575,720B, BPFP=0.3654 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,311,252B, BPFP=0.8323 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,290,676B, BPFP=0.8193 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,387,776B, BPFP=0.8809 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,384,944B, BPFP=0.8791 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,003,856B, BPFP=0.6372 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 969,208B, BPFP=0.6152 +⌛️ [2/4] FRONTEND: Frontend time: 7.978s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.924s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14530003 4.55816275 + layer.9.1 0.14546206 20.76359634 + layer.19.0 0.11891763 15.57170133 + layer.19.1 0.11677460 6.20913481 + layer.29.0 4.29725807 93.10444426 + layer.29.1 4.29692800 48.50238666 + layer.39.0 11.61914761 1544.49561261 + layer.39.1 11.22064282 1476.40331492 + ------------------------------------------------------------------------------------- + TOTAL 3.99505385 401.20104421 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8481844 +BPFP 0.6730 bits/point +EBPFP 0.6730 equivalent bits/point +MSE 401.201044 +---------------------- --------------------------------------------------------- +Time: 22.161s Load: 1.259s, Pack+Encode: 7.978s, Decode+Unpack: 12.924s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 401.2010 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00433-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst (83/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.279s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 513,972B, BPFP=0.3262 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 524,488B, BPFP=0.3329 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,256,064B, BPFP=0.7973 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,220,452B, BPFP=0.7747 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,310,548B, BPFP=0.8319 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,311,308B, BPFP=0.8324 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 965,212B, BPFP=0.6127 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 986,600B, BPFP=0.6262 +⌛️ [2/4] FRONTEND: Frontend time: 7.881s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.781s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67106792 4.60458359 + layer.9.1 2.67195307 4.47738069 + layer.19.0 0.08237472 6.80787585 + layer.19.1 0.08192194 6.48978751 + layer.29.0 0.11152953 131.75510847 + layer.29.1 0.11703055 94.78219451 + layer.39.0 163.01811830 1678.42606435 + layer.39.1 58.15221299 1640.44816380 + ------------------------------------------------------------------------------------- + TOTAL 28.36327613 445.97389485 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8088644 +BPFP 0.6418 bits/point +EBPFP 0.6418 equivalent bits/point +MSE 445.973895 +---------------------- --------------------------------------------------------- +Time: 21.941s Load: 1.279s, Pack+Encode: 7.881s, Decode+Unpack: 12.781s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 445.9739 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00443-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst (84/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.269s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 533,532B, BPFP=0.3387 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 534,152B, BPFP=0.3391 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,263,916B, BPFP=0.8023 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,245,076B, BPFP=0.7903 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,380,004B, BPFP=0.8760 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,382,368B, BPFP=0.8775 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 931,932B, BPFP=0.5915 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 899,756B, BPFP=0.5711 +⌛️ [2/4] FRONTEND: Frontend time: 7.943s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.175s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14620710 4.43945027 + layer.9.1 0.14642976 4.44349553 + layer.19.0 0.11726453 34.32447138 + layer.19.1 0.11958517 11.13173343 + layer.29.0 0.10693079 54.11129956 + layer.29.1 0.10826971 48.30025187 + layer.39.0 43.01306569 1614.24081898 + layer.39.1 17.12450997 1645.81004225 + ------------------------------------------------------------------------------------- + TOTAL 7.61028284 427.10019541 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8170736 +BPFP 0.6483 bits/point +EBPFP 0.6483 equivalent bits/point +MSE 427.100195 +---------------------- --------------------------------------------------------- +Time: 22.388s Load: 1.269s, Pack+Encode: 7.943s, Decode+Unpack: 13.175s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 427.1002 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00444-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst (85/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.278s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 522,200B, BPFP=0.3315 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 509,412B, BPFP=0.3233 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,283,596B, BPFP=0.8148 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,370,896B, BPFP=0.8702 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,283,060B, BPFP=0.8144 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,282,128B, BPFP=0.8138 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 880,648B, BPFP=0.5590 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 909,080B, BPFP=0.5770 +⌛️ [2/4] FRONTEND: Frontend time: 7.880s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.839s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03173930 8.53009920 + layer.9.1 0.03345565 12.54513071 + layer.19.0 3.26068347 6.11915522 + layer.19.1 3.26087326 11.20611594 + layer.29.0 4.24610771 28.55071904 + layer.29.1 4.24089229 28.14710706 + layer.39.0 8.81319124 1305.44450764 + layer.39.1 8.71779153 1278.04281768 + ------------------------------------------------------------------------------------- + TOTAL 4.07559181 334.82320656 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8041020 +BPFP 0.6380 bits/point +EBPFP 0.6380 equivalent bits/point +MSE 334.823207 +---------------------- --------------------------------------------------------- +Time: 21.996s Load: 1.278s, Pack+Encode: 7.880s, Decode+Unpack: 12.839s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 334.8232 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00445-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst (86/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.279s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 529,556B, BPFP=0.3361 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 539,248B, BPFP=0.3423 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,358,504B, BPFP=0.8623 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,346,340B, BPFP=0.8546 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,448,896B, BPFP=0.9197 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,465,896B, BPFP=0.9305 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 945,400B, BPFP=0.6001 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 985,632B, BPFP=0.6256 +⌛️ [2/4] FRONTEND: Frontend time: 7.967s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.749s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00078599 4.47088212 + layer.9.1 0.00079117 4.46985161 + layer.19.0 0.00795310 29.37198621 + layer.19.1 0.00811505 11.18716231 + layer.29.0 4.25797468 64.13362244 + layer.29.1 4.25504309 83.24635400 + layer.39.0 81.06806549 1515.30289243 + layer.39.1 44.82015254 1447.27185570 + ------------------------------------------------------------------------------------- + TOTAL 16.80236014 394.93182585 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8619472 +BPFP 0.6839 bits/point +EBPFP 0.6839 equivalent bits/point +MSE 394.931826 +---------------------- --------------------------------------------------------- +Time: 21.995s Load: 1.279s, Pack+Encode: 7.967s, Decode+Unpack: 12.749s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 394.9318 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00446-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst (87/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.257s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 527,756B, BPFP=0.3350 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 536,356B, BPFP=0.3405 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,364,240B, BPFP=0.8660 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,368,716B, BPFP=0.8688 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,486,092B, BPFP=0.9433 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,489,924B, BPFP=0.9457 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 997,860B, BPFP=0.6334 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 970,780B, BPFP=0.6162 +⌛️ [2/4] FRONTEND: Frontend time: 7.609s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.848s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00082351 4.46848785 + layer.9.1 0.02968625 4.45529394 + layer.19.0 0.00841222 11.20431833 + layer.19.1 0.03743129 20.25897663 + layer.29.0 4.28408194 42.04500376 + layer.29.1 4.28564945 56.72389909 + layer.39.0 8.35370986 1342.54484888 + layer.39.1 8.52557915 1365.44718882 + ------------------------------------------------------------------------------------- + TOTAL 3.19067171 355.89350216 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8741724 +BPFP 0.6936 bits/point +EBPFP 0.6936 equivalent bits/point +MSE 355.893502 +---------------------- --------------------------------------------------------- +Time: 21.714s Load: 1.257s, Pack+Encode: 7.609s, Decode+Unpack: 12.848s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 355.8935 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00447-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst (88/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.247s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 538,500B, BPFP=0.3418 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 530,824B, BPFP=0.3369 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,359,544B, BPFP=0.8630 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,300,016B, BPFP=0.8252 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,393,184B, BPFP=0.8843 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,390,540B, BPFP=0.8826 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 916,148B, BPFP=0.5815 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 903,164B, BPFP=0.5733 +⌛️ [2/4] FRONTEND: Frontend time: 7.823s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.843s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14529990 4.40352089 + layer.9.1 0.14524076 4.54594224 + layer.19.0 0.03780325 39.58206807 + layer.19.1 0.03783790 20.56918265 + layer.29.0 4.32098184 111.85867931 + layer.29.1 4.32100596 116.99276893 + layer.39.0 9.32673680 1549.59993500 + layer.39.1 9.31823369 1647.10529737 + ------------------------------------------------------------------------------------- + TOTAL 3.45664251 436.83217431 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8331920 +BPFP 0.6611 bits/point +EBPFP 0.6611 equivalent bits/point +MSE 436.832174 +---------------------- --------------------------------------------------------- +Time: 21.913s Load: 1.247s, Pack+Encode: 7.823s, Decode+Unpack: 12.843s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 436.8322 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00461-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst (89/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.257s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 550,704B, BPFP=0.3496 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 546,388B, BPFP=0.3468 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,307,728B, BPFP=0.8301 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,250,772B, BPFP=0.7939 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,433,652B, BPFP=0.9100 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,421,700B, BPFP=0.9024 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 951,812B, BPFP=0.6042 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 946,268B, BPFP=0.6006 +⌛️ [2/4] FRONTEND: Frontend time: 7.974s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.771s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14497286 8.61628679 + layer.9.1 0.14497296 4.38451776 + layer.19.0 0.03962668 21.00197534 + layer.19.1 0.11751332 30.03963631 + layer.29.0 0.14529291 135.46013772 + layer.29.1 0.16241527 126.05544158 + layer.39.0 11.40179406 1602.29168021 + layer.39.1 13.03458244 1683.88495288 + ------------------------------------------------------------------------------------- + TOTAL 3.14889631 451.46682857 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8409024 +BPFP 0.6672 bits/point +EBPFP 0.6672 equivalent bits/point +MSE 451.466829 +---------------------- --------------------------------------------------------- +Time: 22.003s Load: 1.257s, Pack+Encode: 7.974s, Decode+Unpack: 12.771s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 451.4668 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00462-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst (90/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.269s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 573,272B, BPFP=0.3639 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 578,460B, BPFP=0.3672 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,299,000B, BPFP=0.8245 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,314,584B, BPFP=0.8344 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,493,192B, BPFP=0.9478 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,495,424B, BPFP=0.9492 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 939,232B, BPFP=0.5962 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 948,588B, BPFP=0.6021 +⌛️ [2/4] FRONTEND: Frontend time: 7.537s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.919s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14513344 4.38230311 + layer.9.1 0.03283094 8.46087022 + layer.19.0 0.11544709 25.58511994 + layer.19.1 0.11326018 34.35628250 + layer.29.0 0.14483232 134.48341526 + layer.29.1 0.14672551 151.44794036 + layer.39.0 10.02784076 1589.80857979 + layer.39.1 15.62606130 1668.65355866 + ------------------------------------------------------------------------------------- + TOTAL 3.29401644 452.14725873 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8641752 +BPFP 0.6857 bits/point +EBPFP 0.6857 equivalent bits/point +MSE 452.147259 +---------------------- --------------------------------------------------------- +Time: 21.725s Load: 1.269s, Pack+Encode: 7.537s, Decode+Unpack: 12.919s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 452.1473 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00463-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst (91/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.234s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 573,176B, BPFP=0.3638 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 583,884B, BPFP=0.3706 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,297,084B, BPFP=0.8233 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,322,316B, BPFP=0.8393 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,520,700B, BPFP=0.9653 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,529,848B, BPFP=0.9711 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 942,684B, BPFP=0.5984 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 967,812B, BPFP=0.6143 +⌛️ [2/4] FRONTEND: Frontend time: 7.747s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.433s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14492303 4.41319511 + layer.9.1 0.14484742 4.38182800 + layer.19.0 0.11740684 56.87797063 + layer.19.1 0.11489933 67.01778315 + layer.29.0 0.12072669 84.02521226 + layer.29.1 0.12118037 129.75798261 + layer.39.0 10.74778980 1681.83295418 + layer.39.1 11.83662176 1710.41225219 + ------------------------------------------------------------------------------------- + TOTAL 2.91854940 467.33989727 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8737504 +BPFP 0.6933 bits/point +EBPFP 0.6933 equivalent bits/point +MSE 467.339897 +---------------------- --------------------------------------------------------- +Time: 21.414s Load: 1.234s, Pack+Encode: 7.747s, Decode+Unpack: 12.433s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 467.3399 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00464-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst (92/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.265s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 598,708B, BPFP=0.3800 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 594,640B, BPFP=0.3774 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,309,900B, BPFP=0.8315 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,294,256B, BPFP=0.8215 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,553,504B, BPFP=0.9861 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,538,316B, BPFP=0.9764 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,023,036B, BPFP=0.6494 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 977,600B, BPFP=0.6205 +⌛️ [2/4] FRONTEND: Frontend time: 7.732s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.065s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14465000 4.37231152 + layer.9.1 0.14489275 4.51758828 + layer.19.0 0.11978787 76.75513386 + layer.19.1 0.12819003 108.85057483 + layer.29.0 0.12519148 138.38208889 + layer.29.1 0.13018718 89.20289649 + layer.39.0 10.77894586 1746.60578486 + layer.39.1 10.25834823 1779.72245694 + ------------------------------------------------------------------------------------- + TOTAL 2.72877418 493.55110446 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8889960 +BPFP 0.7054 bits/point +EBPFP 0.7054 equivalent bits/point +MSE 493.551104 +---------------------- --------------------------------------------------------- +Time: 22.062s Load: 1.265s, Pack+Encode: 7.732s, Decode+Unpack: 13.065s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 493.5511 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00465-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst (93/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.240s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 517,232B, BPFP=0.3283 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 511,668B, BPFP=0.3248 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,196,676B, BPFP=0.7596 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,277,088B, BPFP=0.8106 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,363,324B, BPFP=0.8654 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,371,068B, BPFP=0.8703 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 900,396B, BPFP=0.5715 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 913,508B, BPFP=0.5798 +⌛️ [2/4] FRONTEND: Frontend time: 7.748s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.725s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14553812 4.44183344 + layer.9.1 0.14559401 4.43381401 + layer.19.0 0.04492324 15.32790919 + layer.19.1 0.04213941 15.98373390 + layer.29.0 4.25320263 38.68215795 + layer.29.1 4.25391672 32.65200479 + layer.39.0 8.72311137 1428.63812155 + layer.39.1 8.87262096 1413.21920702 + ------------------------------------------------------------------------------------- + TOTAL 3.31013081 369.17234773 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8050960 +BPFP 0.6388 bits/point +EBPFP 0.6388 equivalent bits/point +MSE 369.172348 +---------------------- --------------------------------------------------------- +Time: 21.712s Load: 1.240s, Pack+Encode: 7.748s, Decode+Unpack: 12.725s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 369.1723 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00471-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst (94/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.259s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 547,476B, BPFP=0.3475 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 546,864B, BPFP=0.3471 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,351,128B, BPFP=0.8576 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,347,192B, BPFP=0.8551 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,462,672B, BPFP=0.9284 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,458,776B, BPFP=0.9260 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 927,520B, BPFP=0.5887 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 921,932B, BPFP=0.5852 +⌛️ [2/4] FRONTEND: Frontend time: 7.911s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.541s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14518122 4.70128137 + layer.9.1 0.14529820 4.55024774 + layer.19.0 0.11833418 43.80605602 + layer.19.1 0.12038008 33.96519032 + layer.29.0 4.31360161 115.54181224 + layer.29.1 4.31792870 192.35964413 + layer.39.0 9.40764201 1464.13747156 + layer.39.1 11.30764416 1573.04614885 + ------------------------------------------------------------------------------------- + TOTAL 3.73450127 429.01348153 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8563560 +BPFP 0.6795 bits/point +EBPFP 0.6795 equivalent bits/point +MSE 429.013482 +---------------------- --------------------------------------------------------- +Time: 21.712s Load: 1.259s, Pack+Encode: 7.911s, Decode+Unpack: 12.541s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 429.0135 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00472-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst (95/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.256s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 575,184B, BPFP=0.3651 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 579,908B, BPFP=0.3681 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,282,908B, BPFP=0.8143 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,288,872B, BPFP=0.8181 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,541,128B, BPFP=0.9782 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,556,500B, BPFP=0.9880 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 971,812B, BPFP=0.6169 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 966,892B, BPFP=0.6137 +⌛️ [2/4] FRONTEND: Frontend time: 7.882s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.572s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00315138 25.12411897 + layer.9.1 0.00505826 21.64060215 + layer.19.0 0.09147678 81.08696579 + layer.19.1 0.09143778 67.42877600 + layer.29.0 0.11015094 58.51378676 + layer.29.1 0.11338039 77.34367383 + layer.39.0 9.14784464 1446.91696458 + layer.39.1 8.98944348 1462.83002925 + ------------------------------------------------------------------------------------- + TOTAL 2.31899296 405.11061467 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8763204 +BPFP 0.6953 bits/point +EBPFP 0.6953 equivalent bits/point +MSE 405.110615 +---------------------- --------------------------------------------------------- +Time: 21.709s Load: 1.256s, Pack+Encode: 7.882s, Decode+Unpack: 12.572s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 405.1106 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00475-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst (96/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.262s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 609,544B, BPFP=0.3869 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 599,760B, BPFP=0.3807 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,335,316B, BPFP=0.8476 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,334,444B, BPFP=0.8470 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,632,208B, BPFP=1.0360 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,615,716B, BPFP=1.0256 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,076,772B, BPFP=0.6835 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,059,848B, BPFP=0.6727 +⌛️ [2/4] FRONTEND: Frontend time: 7.683s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.075s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14591913 8.58647640 + layer.9.1 0.03347605 12.84665588 + layer.19.0 0.12173996 72.17429213 + layer.19.1 0.12099332 53.32473290 + layer.29.0 0.11078974 50.07881053 + layer.29.1 0.11776269 74.99050414 + layer.39.0 10.17800795 1603.12154696 + layer.39.1 9.88744998 1602.07881053 + ------------------------------------------------------------------------------------- + TOTAL 2.58951735 434.65022868 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9263608 +BPFP 0.7350 bits/point +EBPFP 0.7350 equivalent bits/point +MSE 434.650229 +---------------------- --------------------------------------------------------- +Time: 22.021s Load: 1.262s, Pack+Encode: 7.683s, Decode+Unpack: 13.075s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 434.6502 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00476-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst (97/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.284s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 509,992B, BPFP=0.3237 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 511,396B, BPFP=0.3246 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,280,200B, BPFP=0.8126 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,323,376B, BPFP=0.8400 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,492,900B, BPFP=0.9476 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,499,124B, BPFP=0.9516 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 873,476B, BPFP=0.5544 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 889,080B, BPFP=0.5643 +⌛️ [2/4] FRONTEND: Frontend time: 7.839s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.539s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66405266 4.54706066 + layer.9.1 2.66543197 4.53966773 + layer.19.0 3.22131407 6.22688686 + layer.19.1 3.22426883 11.11571869 + layer.29.0 4.27224607 39.43024862 + layer.29.1 4.27784520 53.52715206 + layer.39.0 8.94937744 1398.54809880 + layer.39.1 8.82170070 1331.58319792 + ------------------------------------------------------------------------------------- + TOTAL 4.76202962 356.18975392 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8379544 +BPFP 0.6649 bits/point +EBPFP 0.6649 equivalent bits/point +MSE 356.189754 +---------------------- --------------------------------------------------------- +Time: 21.662s Load: 1.284s, Pack+Encode: 7.839s, Decode+Unpack: 12.539s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 356.1898 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00643-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst (98/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.263s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 544,636B, BPFP=0.3457 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 538,400B, BPFP=0.3417 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,319,560B, BPFP=0.8376 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,296,900B, BPFP=0.8232 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,518,080B, BPFP=0.9636 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,501,500B, BPFP=0.9531 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,036,056B, BPFP=0.6576 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 932,768B, BPFP=0.5921 +⌛️ [2/4] FRONTEND: Frontend time: 7.881s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.162s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00104424 4.49756542 + layer.9.1 0.00091568 8.57642387 + layer.19.0 0.08171424 15.75206801 + layer.19.1 0.08373584 24.81217501 + layer.29.0 4.26071267 28.88835260 + layer.29.1 4.26438533 47.38970588 + layer.39.0 8.39843369 1397.35472863 + layer.39.1 8.51949380 1451.30679233 + ------------------------------------------------------------------------------------- + TOTAL 3.20130444 372.32222647 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8687900 +BPFP 0.6893 bits/point +EBPFP 0.6893 equivalent bits/point +MSE 372.322226 +---------------------- --------------------------------------------------------- +Time: 22.305s Load: 1.263s, Pack+Encode: 7.881s, Decode+Unpack: 13.162s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 372.3222 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00644-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst (99/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.246s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 569,504B, BPFP=0.3615 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 559,732B, BPFP=0.3553 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,262,636B, BPFP=0.8015 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,269,248B, BPFP=0.8057 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,550,412B, BPFP=0.9841 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,549,724B, BPFP=0.9837 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,036,636B, BPFP=0.6580 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,043,364B, BPFP=0.6623 +⌛️ [2/4] FRONTEND: Frontend time: 7.839s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.158s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14622840 12.88466089 + layer.9.1 0.03344178 4.42660547 + layer.19.0 0.12675888 44.61016209 + layer.19.1 0.12382618 67.15048647 + layer.29.0 0.12223263 102.61618460 + layer.29.1 0.12797405 102.09434920 + layer.39.0 10.69978368 1591.08969776 + layer.39.1 8.63538768 1618.17923302 + ------------------------------------------------------------------------------------- + TOTAL 2.50195416 442.88142244 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8841256 +BPFP 0.7015 bits/point +EBPFP 0.7015 equivalent bits/point +MSE 442.881422 +---------------------- --------------------------------------------------------- +Time: 22.243s Load: 1.246s, Pack+Encode: 7.839s, Decode+Unpack: 13.158s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 442.8814 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00272-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst (100/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.275s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 572,072B, BPFP=0.3631 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 573,772B, BPFP=0.3642 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,383,368B, BPFP=0.8781 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,357,452B, BPFP=0.8616 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,437,060B, BPFP=0.9122 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,437,572B, BPFP=0.9125 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 993,736B, BPFP=0.6308 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 978,680B, BPFP=0.6212 +⌛️ [2/4] FRONTEND: Frontend time: 7.822s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.179s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14492994 8.53179588 + layer.9.1 0.14498602 12.47619054 + layer.19.0 0.12957112 15.77051003 + layer.19.1 0.13054295 20.45225666 + layer.29.0 0.16610158 190.02528843 + layer.29.1 0.14872770 135.97762634 + layer.39.0 16.52878844 1593.16428339 + layer.39.1 24.55764797 1714.79915502 + ------------------------------------------------------------------------------------- + TOTAL 5.24391197 461.39963829 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 8733712 +BPFP 0.6930 bits/point +EBPFP 0.6930 equivalent bits/point +MSE 461.399638 +---------------------- --------------------------------------------------------- +Time: 22.276s Load: 1.275s, Pack+Encode: 7.822s, Decode+Unpack: 13.179s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 461.3996 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst + to output-fixed/dinov3-total/lambda0.007/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00278-wflip.zst +------------------------ ---------------------------- +TOTAL PROCESSING SUMMARY +------------------------ ---------------------------- +Total files 100 +Avg BPFP 0.6716 bits/point +Avg EBPFP 0.6716 equivalent bits/point +Avg MSE 417.689147 +Avg Time 21.959s +------------------------ ---------------------------- diff --git a/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log b/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log new file mode 100644 index 0000000000000000000000000000000000000000..f514d0d006b8f50236aa7abe223f7c6e96a473f7 --- /dev/null +++ b/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log @@ -0,0 +1,15744 @@ +Experiment: dtufc_hyperprior-featurecoding_dinov3-total_individual +Log file: output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log +DTUFCCodecConfig: + arch: hyperprior-featurecoding + handler: dinov3-total + checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.01_epochs600_lr0.0001_bs360_patch256-256_checkpoint_best.pth.tar + transform_type: kmeans + transform_mapping:featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json + bit_depth: 8 + device: cuda:0 +Loading checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.01_epochs600_lr0.0001_bs360_patch256-256_checkpoint_best.pth.tar +Checkpoint epoch: 599 +Loaded hyperprior-featurecoding (1-channel) on cuda:0 +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_9_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.9' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_9_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_19_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.19' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_19_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_29_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.29' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_29_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_39_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.39' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_39_0.json +Loaded per-key mappings: model=dinov3-total + Keys: ['layer.9', 'layer.19', 'layer.29', 'layer.39'] +---------------- ----------------------------------------------------------------------------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +Checkpoint codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.01_epochs600_lr0.0001_bs360_patch256-256_checkpoint_best.pth.tar +Transform type kmeans +Transform config featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json +Input ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features +Output output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test +---------------- ----------------------------------------------------------------------------------------------------------------------------- +Files found: 100 +---------------------------------------------------------------------- + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst (1/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.257s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 738,680B, BPFP=0.4689 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 736,120B, BPFP=0.4673 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,452,936B, BPFP=0.9223 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,463,068B, BPFP=0.9287 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,609,416B, BPFP=1.0216 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,619,880B, BPFP=1.0282 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,073,672B, BPFP=0.6815 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,084,428B, BPFP=0.6883 +⌛️ [2/4] FRONTEND: Frontend time: 7.681s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.552s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14561056 20.92240057 + layer.9.1 0.14522085 16.55254458 + layer.19.0 3.25142184 25.86382079 + layer.19.1 3.25206135 34.96406809 + layer.29.0 4.23946030 60.06083442 + layer.29.1 4.24539299 45.09964048 + layer.39.0 32.17105490 1998.93678908 + layer.39.1 19.15684032 1916.09229769 + ------------------------------------------------------------------------------------- + TOTAL 8.32588289 514.81154946 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9778200 +BPFP 0.7758 bits/point +EBPFP 0.7758 equivalent bits/point +MSE 514.811549 +---------------------- --------------------------------------------------------- +Time: 21.491s Load: 1.257s, Pack+Encode: 7.681s, Decode+Unpack: 12.552s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 514.8115 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00512-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst (2/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.207s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 862,108B, BPFP=0.5472 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 847,392B, BPFP=0.5379 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,667,984B, BPFP=1.0588 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,659,480B, BPFP=1.0534 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,858,304B, BPFP=1.1796 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,843,240B, BPFP=1.1700 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,230,060B, BPFP=0.7808 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,225,768B, BPFP=0.7781 +⌛️ [2/4] FRONTEND: Frontend time: 7.040s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.524s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03307773 8.63431241 + layer.9.1 0.03291117 16.74306599 + layer.19.0 0.04156009 48.96862305 + layer.19.1 0.03760627 104.19988016 + layer.29.0 4.28582750 81.93200561 + layer.29.1 4.28551552 110.81761862 + layer.39.0 9.83402183 1733.59733507 + layer.39.1 9.85397836 1812.83912902 + ------------------------------------------------------------------------------------- + TOTAL 3.55056231 489.71649624 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11194336 +BPFP 0.8882 bits/point +EBPFP 0.8882 equivalent bits/point +MSE 489.716496 +---------------------- --------------------------------------------------------- +Time: 20.771s Load: 1.207s, Pack+Encode: 7.040s, Decode+Unpack: 12.524s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 489.7165 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00657-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst (3/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 914,216B, BPFP=0.5803 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 911,916B, BPFP=0.5788 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,668,260B, BPFP=1.0589 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,670,308B, BPFP=1.0602 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,874,340B, BPFP=1.1897 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,876,828B, BPFP=1.1913 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,165,688B, BPFP=0.7399 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,173,812B, BPFP=0.7451 +⌛️ [2/4] FRONTEND: Frontend time: 6.985s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.574s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00083900 29.31559250 + layer.9.1 0.00259629 33.57523054 + layer.19.0 0.00955961 80.84396835 + layer.19.1 0.08538111 108.62674683 + layer.29.0 0.11631418 199.73033799 + layer.29.1 0.11200302 140.05243541 + layer.39.0 14.47657393 2278.30857979 + layer.39.1 13.08093694 2211.97123822 + ------------------------------------------------------------------------------------- + TOTAL 3.48552551 635.30301620 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11255368 +BPFP 0.8930 bits/point +EBPFP 0.8930 equivalent bits/point +MSE 635.303016 +---------------------- --------------------------------------------------------- +Time: 20.755s Load: 1.196s, Pack+Encode: 6.985s, Decode+Unpack: 12.574s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 635.3030 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00676-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst (4/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.198s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 761,296B, BPFP=0.4832 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 763,240B, BPFP=0.4845 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,510,880B, BPFP=0.9590 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,512,724B, BPFP=0.9602 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,639,316B, BPFP=1.0406 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,636,404B, BPFP=1.0387 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,128,968B, BPFP=0.7166 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,137,740B, BPFP=0.7222 +⌛️ [2/4] FRONTEND: Frontend time: 6.961s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.556s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.70169554 8.43364834 + layer.9.1 0.03294074 24.98169636 + layer.19.0 3.25671692 53.98564450 + layer.19.1 3.25834093 30.58502092 + layer.29.0 0.10810242 216.82645434 + layer.29.1 0.10661203 173.50170621 + layer.39.0 8.95005916 1617.19385765 + layer.39.1 8.98756017 1546.23074423 + ------------------------------------------------------------------------------------- + TOTAL 3.42525349 458.96734657 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10090568 +BPFP 0.8006 bits/point +EBPFP 0.8006 equivalent bits/point +MSE 458.967347 +---------------------- --------------------------------------------------------- +Time: 20.716s Load: 1.198s, Pack+Encode: 6.961s, Decode+Unpack: 12.556s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 458.9673 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00688-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst (5/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 860,968B, BPFP=0.5465 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 859,380B, BPFP=0.5455 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,610,812B, BPFP=1.0225 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,612,988B, BPFP=1.0238 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,798,504B, BPFP=1.1416 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,805,152B, BPFP=1.1458 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,123,824B, BPFP=0.7133 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,184,044B, BPFP=0.7516 +⌛️ [2/4] FRONTEND: Frontend time: 6.991s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.636s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14527870 16.76638924 + layer.9.1 0.14521496 17.17268367 + layer.19.0 0.03964342 90.45642062 + layer.19.1 0.03956446 67.35170722 + layer.29.0 0.12258449 217.08610253 + layer.29.1 0.12735008 260.33518850 + layer.39.0 32.94776263 2042.42638934 + layer.39.1 29.25669534 2155.48521287 + ------------------------------------------------------------------------------------- + TOTAL 7.85301176 608.38501175 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10855672 +BPFP 0.8613 bits/point +EBPFP 0.8613 equivalent bits/point +MSE 608.385012 +---------------------- --------------------------------------------------------- +Time: 20.821s Load: 1.193s, Pack+Encode: 6.991s, Decode+Unpack: 12.636s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 608.3850 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00693-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst (6/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.191s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 793,728B, BPFP=0.5038 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 800,436B, BPFP=0.5081 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,506,520B, BPFP=0.9563 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,504,340B, BPFP=0.9549 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,633,832B, BPFP=1.0371 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,632,916B, BPFP=1.0365 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,092,048B, BPFP=0.6932 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,082,772B, BPFP=0.6873 +⌛️ [2/4] FRONTEND: Frontend time: 7.004s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.571s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66816044 16.71629225 + layer.9.1 2.66817504 33.41698489 + layer.19.0 3.22262959 17.79475747 + layer.19.1 3.22037432 35.95062662 + layer.29.0 4.30448692 261.40296961 + layer.29.1 4.31085282 198.59223676 + layer.39.0 38.33931691 1677.29574261 + layer.39.1 57.25219370 1796.02079948 + ------------------------------------------------------------------------------------- + TOTAL 14.49827372 504.64880121 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10046592 +BPFP 0.7971 bits/point +EBPFP 0.7971 equivalent bits/point +MSE 504.648801 +---------------------- --------------------------------------------------------- +Time: 20.765s Load: 1.191s, Pack+Encode: 7.004s, Decode+Unpack: 12.571s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 504.6488 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00731-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst (7/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 877,716B, BPFP=0.5571 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 872,880B, BPFP=0.5541 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,588,892B, BPFP=1.0085 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,582,084B, BPFP=1.0042 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,726,420B, BPFP=1.0958 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,719,252B, BPFP=1.0913 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,096,276B, BPFP=0.6959 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,100,888B, BPFP=0.6988 +⌛️ [2/4] FRONTEND: Frontend time: 6.996s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.535s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11332920 12.57080380 + layer.9.1 0.00092169 16.67529122 + layer.19.0 3.23006092 34.81396754 + layer.19.1 3.23257961 34.81442710 + layer.29.0 4.28548854 130.49209864 + layer.29.1 4.27808990 86.81921311 + layer.39.0 10.57841825 1667.25674358 + layer.39.1 20.33118703 1662.98163796 + ------------------------------------------------------------------------------------- + TOTAL 5.75625939 455.80302287 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10564408 +BPFP 0.8382 bits/point +EBPFP 0.8382 equivalent bits/point +MSE 455.803023 +---------------------- --------------------------------------------------------- +Time: 20.726s Load: 1.196s, Pack+Encode: 6.996s, Decode+Unpack: 12.535s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 455.8030 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00061-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst (8/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.203s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 913,592B, BPFP=0.5799 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 918,196B, BPFP=0.5828 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,598,204B, BPFP=1.0145 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,605,744B, BPFP=1.0192 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,676,464B, BPFP=1.0641 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,693,700B, BPFP=1.0751 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,120,792B, BPFP=0.7114 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,127,124B, BPFP=0.7154 +⌛️ [2/4] FRONTEND: Frontend time: 6.993s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.533s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14443954 49.48498436 + layer.9.1 0.14435121 49.68997299 + layer.19.0 0.03807715 73.86425496 + layer.19.1 0.03781311 67.75004062 + layer.29.0 0.10781899 109.35602860 + layer.29.1 0.10618912 54.19330415 + layer.39.0 9.30898666 1612.68719532 + layer.39.1 9.83625107 1629.32564186 + ------------------------------------------------------------------------------------- + TOTAL 2.46549086 455.79392786 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10653816 +BPFP 0.8453 bits/point +EBPFP 0.8453 equivalent bits/point +MSE 455.793928 +---------------------- --------------------------------------------------------- +Time: 20.729s Load: 1.203s, Pack+Encode: 6.993s, Decode+Unpack: 12.533s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 455.7939 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00175-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst (9/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.192s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 950,284B, BPFP=0.6032 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 944,820B, BPFP=0.5997 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,683,884B, BPFP=1.0688 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,680,776B, BPFP=1.0669 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,870,324B, BPFP=1.1872 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,862,808B, BPFP=1.1824 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,253,964B, BPFP=0.7960 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,235,780B, BPFP=0.7844 +⌛️ [2/4] FRONTEND: Frontend time: 7.216s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.654s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14543928 29.31191349 + layer.9.1 0.14562574 21.46293569 + layer.19.0 0.11552505 82.66767042 + layer.19.1 0.12052174 68.21206735 + layer.29.0 0.10841144 92.73403477 + layer.29.1 0.10845811 103.21774456 + layer.39.0 9.17501701 1763.42703932 + layer.39.1 9.20635778 1771.44800130 + ------------------------------------------------------------------------------------- + TOTAL 2.39066952 491.56017586 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11482640 +BPFP 0.9111 bits/point +EBPFP 0.9111 equivalent bits/point +MSE 491.560176 +---------------------- --------------------------------------------------------- +Time: 21.062s Load: 1.192s, Pack+Encode: 7.216s, Decode+Unpack: 12.654s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 491.5602 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00186-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst (10/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.187s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 844,772B, BPFP=0.5362 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 841,452B, BPFP=0.5341 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,460,540B, BPFP=0.9271 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,459,116B, BPFP=0.9262 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,515,368B, BPFP=0.9619 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,526,048B, BPFP=0.9687 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,006,756B, BPFP=0.6390 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,031,324B, BPFP=0.6546 +⌛️ [2/4] FRONTEND: Frontend time: 6.944s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.396s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.78226592 54.10340835 + layer.9.1 2.78427046 49.51646795 + layer.19.0 3.22580366 27.01578750 + layer.19.1 3.22969594 44.64207223 + layer.29.0 4.29525448 113.86526040 + layer.29.1 0.11349234 109.43911480 + layer.39.0 8.89338553 1588.03217420 + layer.39.1 8.88767087 1593.32206695 + ------------------------------------------------------------------------------------- + TOTAL 4.27647990 447.49204405 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9685376 +BPFP 0.7685 bits/point +EBPFP 0.7685 equivalent bits/point +MSE 447.492044 +---------------------- --------------------------------------------------------- +Time: 20.526s Load: 1.187s, Pack+Encode: 6.944s, Decode+Unpack: 12.396s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 447.4920 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00282-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst (11/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 830,364B, BPFP=0.5271 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 813,432B, BPFP=0.5163 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,510,744B, BPFP=0.9589 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,489,864B, BPFP=0.9457 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,584,436B, BPFP=1.0057 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,574,944B, BPFP=0.9997 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,013,396B, BPFP=0.6433 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,021,684B, BPFP=0.6485 +⌛️ [2/4] FRONTEND: Frontend time: 7.225s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.701s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14514284 57.77146978 + layer.9.1 0.14518188 37.63916254 + layer.19.0 0.04057091 78.20128676 + layer.19.1 0.04041447 83.44199911 + layer.29.0 4.25641542 64.64006134 + layer.29.1 4.26613502 61.03414446 + layer.39.0 12.58558458 1724.63243419 + layer.39.1 8.96866240 1692.81524212 + ------------------------------------------------------------------------------------- + TOTAL 3.80601344 475.02197504 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9838864 +BPFP 0.7807 bits/point +EBPFP 0.7807 equivalent bits/point +MSE 475.021975 +---------------------- --------------------------------------------------------- +Time: 21.122s Load: 1.196s, Pack+Encode: 7.225s, Decode+Unpack: 12.701s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 475.0220 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00524-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst (12/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 836,048B, BPFP=0.5307 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 839,076B, BPFP=0.5326 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,557,316B, BPFP=0.9885 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,560,028B, BPFP=0.9902 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,701,140B, BPFP=1.0798 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,711,420B, BPFP=1.0863 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,101,932B, BPFP=0.6995 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,129,924B, BPFP=0.7172 +⌛️ [2/4] FRONTEND: Frontend time: 6.996s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.638s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67015388 8.91998852 + layer.9.1 0.00076871 8.51110812 + layer.19.0 3.22151687 6.94755190 + layer.19.1 3.22388957 20.89352200 + layer.29.0 4.24084786 46.86015701 + layer.29.1 4.24602234 57.42603388 + layer.39.0 7.87160790 1510.24861878 + layer.39.1 9.85764150 1527.21530712 + ------------------------------------------------------------------------------------- + TOTAL 4.41655608 398.37778592 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10436884 +BPFP 0.8281 bits/point +EBPFP 0.8281 equivalent bits/point +MSE 398.377786 +---------------------- --------------------------------------------------------- +Time: 20.827s Load: 1.193s, Pack+Encode: 6.996s, Decode+Unpack: 12.638s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 398.3778 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00538-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst (13/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 937,672B, BPFP=0.5952 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 946,660B, BPFP=0.6009 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,648,704B, BPFP=1.0465 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,653,540B, BPFP=1.0496 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,821,020B, BPFP=1.1559 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,819,988B, BPFP=1.1552 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,179,992B, BPFP=0.7490 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,164,096B, BPFP=0.7389 +⌛️ [2/4] FRONTEND: Frontend time: 6.982s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.540s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00068237 29.11480338 + layer.9.1 0.00070576 12.91135085 + layer.19.0 0.00823322 50.30265884 + layer.19.1 0.08594799 101.31322108 + layer.29.0 0.12200666 218.88105297 + layer.29.1 0.12451052 243.07513406 + layer.39.0 55.99513528 2237.76584335 + layer.39.1 28.81185256 2228.72651934 + ------------------------------------------------------------------------------------- + TOTAL 10.64363429 640.26132298 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11171672 +BPFP 0.8864 bits/point +EBPFP 0.8864 equivalent bits/point +MSE 640.261323 +---------------------- --------------------------------------------------------- +Time: 20.716s Load: 1.193s, Pack+Encode: 6.982s, Decode+Unpack: 12.540s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 640.2613 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00927-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst (14/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.200s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 899,116B, BPFP=0.5707 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 894,800B, BPFP=0.5680 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,553,052B, BPFP=0.9858 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,562,888B, BPFP=0.9920 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,632,212B, BPFP=1.0360 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,661,472B, BPFP=1.0546 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,104,380B, BPFP=0.7010 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,107,444B, BPFP=0.7029 +⌛️ [2/4] FRONTEND: Frontend time: 7.054s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.625s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14556937 49.47064917 + layer.9.1 0.03327741 70.74982735 + layer.19.0 0.11590617 86.20150512 + layer.19.1 0.11733878 156.32076698 + layer.29.0 0.11334742 81.89670844 + layer.29.1 4.29039579 130.94325845 + layer.39.0 9.10722066 1635.72781930 + layer.39.1 44.52401893 1738.45206370 + ------------------------------------------------------------------------------------- + TOTAL 7.30588432 493.72032481 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10415364 +BPFP 0.8264 bits/point +EBPFP 0.8264 equivalent bits/point +MSE 493.720325 +---------------------- --------------------------------------------------------- +Time: 20.880s Load: 1.200s, Pack+Encode: 7.054s, Decode+Unpack: 12.625s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 493.7203 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00933-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst (15/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,009,900B, BPFP=0.6410 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,013,028B, BPFP=0.6430 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,643,564B, BPFP=1.0433 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,648,972B, BPFP=1.0467 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,747,704B, BPFP=1.1094 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,758,332B, BPFP=1.1161 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,032,852B, BPFP=0.6556 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,055,640B, BPFP=0.6701 +⌛️ [2/4] FRONTEND: Frontend time: 6.996s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.603s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11318631 71.21409855 + layer.9.1 0.11319129 46.71730277 + layer.19.0 0.00665199 30.95734228 + layer.19.1 0.00853768 7.12555731 + layer.29.0 4.27225940 56.84068289 + layer.29.1 4.27324961 82.10797855 + layer.39.0 14.80262837 2000.17192070 + layer.39.1 16.56649765 2020.58498538 + ------------------------------------------------------------------------------------- + TOTAL 5.01952529 539.46498355 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10909992 +BPFP 0.8656 bits/point +EBPFP 0.8656 equivalent bits/point +MSE 539.464984 +---------------------- --------------------------------------------------------- +Time: 20.792s Load: 1.193s, Pack+Encode: 6.996s, Decode+Unpack: 12.603s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 539.4650 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01095-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst (16/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.190s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 906,244B, BPFP=0.5752 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 904,880B, BPFP=0.5744 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,583,200B, BPFP=1.0049 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,584,344B, BPFP=1.0057 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,708,008B, BPFP=1.0842 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,702,432B, BPFP=1.0806 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,100,476B, BPFP=0.6985 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,077,572B, BPFP=0.6840 +⌛️ [2/4] FRONTEND: Frontend time: 7.023s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.640s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00065117 12.75092547 + layer.9.1 0.00066201 12.79758084 + layer.19.0 0.00984582 26.15316512 + layer.19.1 0.01156107 30.14865839 + layer.29.0 4.26547583 86.36927202 + layer.29.1 4.26296603 91.18362041 + layer.39.0 11.21169412 1756.06109847 + layer.39.1 9.31977106 1749.43857654 + ------------------------------------------------------------------------------------- + TOTAL 3.63532839 470.61286216 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10567156 +BPFP 0.8384 bits/point +EBPFP 0.8384 equivalent bits/point +MSE 470.612862 +---------------------- --------------------------------------------------------- +Time: 20.853s Load: 1.190s, Pack+Encode: 7.023s, Decode+Unpack: 12.640s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 470.6129 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01125-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst (17/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.200s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 861,716B, BPFP=0.5470 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 858,724B, BPFP=0.5451 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,505,132B, BPFP=0.9554 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,505,260B, BPFP=0.9555 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,558,304B, BPFP=0.9891 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,556,716B, BPFP=0.9881 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,030,648B, BPFP=0.6542 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,013,580B, BPFP=0.6434 +⌛️ [2/4] FRONTEND: Frontend time: 7.007s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.497s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00085355 8.57375792 + layer.9.1 0.00085581 8.99047875 + layer.19.0 0.00808159 30.79417604 + layer.19.1 0.00635426 21.46233649 + layer.29.0 4.24551200 100.15761090 + layer.29.1 4.24803037 53.64547449 + layer.39.0 9.19283951 1666.70978226 + layer.39.1 9.46657027 1673.19174521 + ------------------------------------------------------------------------------------- + TOTAL 3.39613717 445.44067026 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9890080 +BPFP 0.7847 bits/point +EBPFP 0.7847 equivalent bits/point +MSE 445.440670 +---------------------- --------------------------------------------------------- +Time: 20.704s Load: 1.200s, Pack+Encode: 7.007s, Decode+Unpack: 12.497s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 445.4407 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01135-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst (18/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.195s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 845,040B, BPFP=0.5364 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 846,764B, BPFP=0.5375 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,616,088B, BPFP=1.0258 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,617,268B, BPFP=1.0266 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,800,048B, BPFP=1.1426 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,804,652B, BPFP=1.1455 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,187,872B, BPFP=0.7540 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,159,236B, BPFP=0.7358 +⌛️ [2/4] FRONTEND: Frontend time: 7.258s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.471s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67056234 8.80306635 + layer.9.1 2.67147828 24.87579725 + layer.19.0 0.00618387 48.46547977 + layer.19.1 0.08383032 90.28730297 + layer.29.0 4.28489822 67.71889726 + layer.29.1 4.28470970 97.79602697 + layer.39.0 10.15376305 1676.30175496 + layer.39.1 8.47863686 1719.34839129 + ------------------------------------------------------------------------------------- + TOTAL 4.07925783 466.69958960 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10876968 +BPFP 0.8630 bits/point +EBPFP 0.8630 equivalent bits/point +MSE 466.699590 +---------------------- --------------------------------------------------------- +Time: 20.924s Load: 1.195s, Pack+Encode: 7.258s, Decode+Unpack: 12.471s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 466.6996 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01161-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst (19/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 834,668B, BPFP=0.5298 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 841,380B, BPFP=0.5341 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,590,028B, BPFP=1.0093 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,595,800B, BPFP=1.0129 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,752,944B, BPFP=1.1127 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,765,100B, BPFP=1.1204 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,121,952B, BPFP=0.7122 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,109,140B, BPFP=0.7040 +⌛️ [2/4] FRONTEND: Frontend time: 7.036s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.751s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67132404 29.25566451 + layer.9.1 2.67117709 12.74588555 + layer.19.0 0.00597838 77.03457101 + layer.19.1 0.00605309 44.21833360 + layer.29.0 4.29273040 154.21709457 + layer.29.1 4.29206328 115.91365372 + layer.39.0 9.96127074 1982.70376991 + layer.39.1 10.21295854 1917.47318817 + ------------------------------------------------------------------------------------- + TOTAL 4.26419445 541.69527013 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10611012 +BPFP 0.8419 bits/point +EBPFP 0.8419 equivalent bits/point +MSE 541.695270 +---------------------- --------------------------------------------------------- +Time: 20.983s Load: 1.196s, Pack+Encode: 7.036s, Decode+Unpack: 12.751s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 541.6953 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01162-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst (20/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 839,892B, BPFP=0.5331 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 840,784B, BPFP=0.5337 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,506,368B, BPFP=0.9562 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,509,444B, BPFP=0.9581 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,558,364B, BPFP=0.9892 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,556,368B, BPFP=0.9879 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,074,504B, BPFP=0.6820 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,048,048B, BPFP=0.6652 +⌛️ [2/4] FRONTEND: Frontend time: 6.988s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.657s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14568686 8.55979470 + layer.9.1 0.14558674 12.73359045 + layer.19.0 0.00960369 68.61256906 + layer.19.1 0.03847206 31.54957903 + layer.29.0 4.24438723 68.49601377 + layer.29.1 4.24578970 97.72356394 + layer.39.0 9.23757985 1693.97481313 + layer.39.1 9.43674592 1722.34091648 + ------------------------------------------------------------------------------------- + TOTAL 3.43798151 462.99885507 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9933772 +BPFP 0.7882 bits/point +EBPFP 0.7882 equivalent bits/point +MSE 462.998855 +---------------------- --------------------------------------------------------- +Time: 20.841s Load: 1.196s, Pack+Encode: 6.988s, Decode+Unpack: 12.657s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 462.9989 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01163-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst (21/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.198s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 945,464B, BPFP=0.6001 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 942,204B, BPFP=0.5981 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,694,788B, BPFP=1.0758 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,687,204B, BPFP=1.0710 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,861,008B, BPFP=1.1813 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,849,200B, BPFP=1.1738 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,176,716B, BPFP=0.7469 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,154,112B, BPFP=0.7326 +⌛️ [2/4] FRONTEND: Frontend time: 7.005s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.582s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11307967 25.13229962 + layer.9.1 0.00073224 21.63708056 + layer.19.0 0.08207503 72.78223513 + layer.19.1 0.08214869 81.24582589 + layer.29.0 4.26728487 71.06872055 + layer.29.1 4.26774951 85.66671068 + layer.39.0 12.81553410 1812.34400390 + layer.39.1 23.05196315 1856.47367566 + ------------------------------------------------------------------------------------- + TOTAL 5.58507091 503.29381900 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11310696 +BPFP 0.8974 bits/point +EBPFP 0.8974 equivalent bits/point +MSE 503.293819 +---------------------- --------------------------------------------------------- +Time: 20.785s Load: 1.198s, Pack+Encode: 7.005s, Decode+Unpack: 12.582s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 503.2938 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01195-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst (22/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.200s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,138,708B, BPFP=0.7228 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,139,304B, BPFP=0.7232 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,807,648B, BPFP=1.1474 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,814,172B, BPFP=1.1515 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,022,988B, BPFP=1.2841 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,028,600B, BPFP=1.2877 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,221,528B, BPFP=0.7754 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,215,700B, BPFP=0.7717 +⌛️ [2/4] FRONTEND: Frontend time: 7.109s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.478s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14519844 234.68221888 + layer.9.1 0.14499054 231.01811830 + layer.19.0 0.12156012 236.11293468 + layer.19.1 0.12030756 189.87961082 + layer.29.0 0.12020218 144.61494556 + layer.29.1 0.12115470 134.06822798 + layer.39.0 8.85439666 1887.11683458 + layer.39.1 8.75438231 1897.51543711 + ------------------------------------------------------------------------------------- + TOTAL 2.29777406 619.37604099 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 12388648 +BPFP 0.9830 bits/point +EBPFP 0.9830 equivalent bits/point +MSE 619.376041 +---------------------- --------------------------------------------------------- +Time: 20.787s Load: 1.200s, Pack+Encode: 7.109s, Decode+Unpack: 12.478s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 619.3760 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00083-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst (23/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,145,776B, BPFP=0.7273 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,142,356B, BPFP=0.7251 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,800,596B, BPFP=1.1429 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,791,608B, BPFP=1.1372 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,990,844B, BPFP=1.2637 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,974,752B, BPFP=1.2535 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,151,224B, BPFP=0.7307 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,152,520B, BPFP=0.7316 +⌛️ [2/4] FRONTEND: Frontend time: 7.026s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.609s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14541288 239.74280955 + layer.9.1 0.14479464 219.09877722 + layer.19.0 0.11855170 254.87662496 + layer.19.1 0.11778439 325.29151771 + layer.29.0 0.12648388 142.23219654 + layer.29.1 0.12520221 158.27827429 + layer.39.0 8.37129624 1838.28355541 + layer.39.1 8.45478741 1826.43695158 + ------------------------------------------------------------------------------------- + TOTAL 2.20053917 625.53008841 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 12149676 +BPFP 0.9640 bits/point +EBPFP 0.9640 equivalent bits/point +MSE 625.530088 +---------------------- --------------------------------------------------------- +Time: 20.828s Load: 1.193s, Pack+Encode: 7.026s, Decode+Unpack: 12.609s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 625.5301 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00084-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst (24/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,182,888B, BPFP=0.7508 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,179,836B, BPFP=0.7489 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,884,664B, BPFP=1.1963 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,883,256B, BPFP=1.1954 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,119,396B, BPFP=1.3453 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,118,880B, BPFP=1.3450 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,381,892B, BPFP=0.8772 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,382,592B, BPFP=0.8776 +⌛️ [2/4] FRONTEND: Frontend time: 7.289s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.651s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14458197 182.46613991 + layer.9.1 0.14461228 161.00151324 + layer.19.0 0.12127609 189.82009669 + layer.19.1 0.12505172 198.11047693 + layer.29.0 0.11568762 103.72483344 + layer.29.1 0.11796058 114.47411237 + layer.39.0 8.63782956 1821.83945401 + layer.39.1 8.69862780 2004.97741306 + ------------------------------------------------------------------------------------- + TOTAL 2.26320345 597.05175496 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 13133404 +BPFP 1.0421 bits/point +EBPFP 1.0421 equivalent bits/point +MSE 597.051755 +---------------------- --------------------------------------------------------- +Time: 21.133s Load: 1.193s, Pack+Encode: 7.289s, Decode+Unpack: 12.651s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 597.0518 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00085-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst (25/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.194s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,174,492B, BPFP=0.7455 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,178,652B, BPFP=0.7481 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,888,692B, BPFP=1.1988 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,889,196B, BPFP=1.1992 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,121,180B, BPFP=1.3464 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,121,096B, BPFP=1.3464 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,422,580B, BPFP=0.9030 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,425,648B, BPFP=0.9049 +⌛️ [2/4] FRONTEND: Frontend time: 7.470s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.785s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14478087 128.11753534 + layer.9.1 0.14472154 116.64676430 + layer.19.0 0.13423899 291.26856516 + layer.19.1 0.13534726 239.79446295 + layer.29.0 0.11251127 118.80040421 + layer.29.1 0.11242151 103.44352251 + layer.39.0 10.58490794 1943.12918427 + layer.39.1 8.80008176 1928.87374066 + ------------------------------------------------------------------------------------- + TOTAL 2.52112639 608.75927242 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 13221536 +BPFP 1.0490 bits/point +EBPFP 1.0490 equivalent bits/point +MSE 608.759272 +---------------------- --------------------------------------------------------- +Time: 21.449s Load: 1.194s, Pack+Encode: 7.470s, Decode+Unpack: 12.785s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 608.7593 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00086-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst (26/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.197s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,087,236B, BPFP=0.6901 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,087,932B, BPFP=0.6906 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,763,784B, BPFP=1.1196 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,764,112B, BPFP=1.1198 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,963,412B, BPFP=1.2463 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,963,708B, BPFP=1.2465 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,282,352B, BPFP=0.8140 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,271,340B, BPFP=0.8070 +⌛️ [2/4] FRONTEND: Frontend time: 7.162s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.712s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14590075 178.21648521 + layer.9.1 0.14620647 217.64368703 + layer.19.0 0.11628058 109.74239316 + layer.19.1 0.11601873 184.53398196 + layer.29.0 0.11558260 160.32133572 + layer.29.1 0.11828149 165.01929639 + layer.39.0 28.43028163 2170.61033474 + layer.39.1 24.81181701 2104.15160871 + ------------------------------------------------------------------------------------- + TOTAL 6.75004616 661.27989037 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 12183876 +BPFP 0.9667 bits/point +EBPFP 0.9667 equivalent bits/point +MSE 661.279890 +---------------------- --------------------------------------------------------- +Time: 21.071s Load: 1.197s, Pack+Encode: 7.162s, Decode+Unpack: 12.712s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 661.2799 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00090-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst (27/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,038,284B, BPFP=0.6591 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,035,892B, BPFP=0.6575 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,799,200B, BPFP=1.1420 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,793,604B, BPFP=1.1385 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,033,412B, BPFP=1.2907 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,023,140B, BPFP=1.2842 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,326,296B, BPFP=0.8419 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,303,616B, BPFP=0.8275 +⌛️ [2/4] FRONTEND: Frontend time: 7.055s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.787s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11529700 42.10086935 + layer.9.1 0.14629077 41.93087575 + layer.19.0 0.09721754 77.48504022 + layer.19.1 0.12446257 123.50476316 + layer.29.0 4.28687864 65.78042736 + layer.29.1 4.28715508 101.11245734 + layer.39.0 11.34089363 1799.94458889 + layer.39.1 19.75513766 1918.43548911 + ------------------------------------------------------------------------------------- + TOTAL 5.01916661 521.28681390 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 12353444 +BPFP 0.9802 bits/point +EBPFP 0.9802 equivalent bits/point +MSE 521.286814 +---------------------- --------------------------------------------------------- +Time: 21.038s Load: 1.196s, Pack+Encode: 7.055s, Decode+Unpack: 12.787s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 521.2868 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00116-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst (28/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.191s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 995,880B, BPFP=0.6321 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 982,356B, BPFP=0.6236 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,710,968B, BPFP=1.0860 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,714,352B, BPFP=1.0882 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,912,556B, BPFP=1.2140 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,921,764B, BPFP=1.2198 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,291,368B, BPFP=0.8197 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,296,996B, BPFP=0.8233 +⌛️ [2/4] FRONTEND: Frontend time: 7.146s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.928s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14538718 62.34701007 + layer.9.1 0.14538559 66.77145962 + layer.19.0 0.11434236 124.01993622 + layer.19.1 0.11406084 207.59741631 + layer.29.0 0.11219077 119.49321579 + layer.29.1 0.11281304 133.53654127 + layer.39.0 79.88316542 2160.53623659 + layer.39.1 46.71980622 2195.63698408 + ------------------------------------------------------------------------------------- + TOTAL 15.91839393 633.74235000 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11826240 +BPFP 0.9383 bits/point +EBPFP 0.9383 equivalent bits/point +MSE 633.742350 +---------------------- --------------------------------------------------------- +Time: 21.266s Load: 1.191s, Pack+Encode: 7.146s, Decode+Unpack: 12.928s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 633.7423 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00314-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst (29/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.209s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,078,500B, BPFP=0.6846 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,077,400B, BPFP=0.6839 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,741,380B, BPFP=1.1053 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,737,480B, BPFP=1.1029 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,949,148B, BPFP=1.2372 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,943,412B, BPFP=1.2336 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,323,600B, BPFP=0.8402 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,317,936B, BPFP=0.8366 +⌛️ [2/4] FRONTEND: Frontend time: 7.107s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.746s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14512126 161.26719410 + layer.9.1 0.14517278 112.74142834 + layer.19.0 0.11689420 213.72046636 + layer.19.1 0.12099910 295.67569873 + layer.29.0 0.11847120 115.23245044 + layer.29.1 0.12399357 163.98459335 + layer.39.0 75.86630139 2262.88300292 + layer.39.1 56.61936342 2120.39876503 + ------------------------------------------------------------------------------------- + TOTAL 16.65703962 680.73794991 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 12168856 +BPFP 0.9655 bits/point +EBPFP 0.9655 equivalent bits/point +MSE 680.737950 +---------------------- --------------------------------------------------------- +Time: 21.062s Load: 1.209s, Pack+Encode: 7.107s, Decode+Unpack: 12.746s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 680.7379 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00316-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst (30/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.194s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,013,908B, BPFP=0.6436 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,011,608B, BPFP=0.6421 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,755,424B, BPFP=1.1143 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,748,980B, BPFP=1.1102 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,950,320B, BPFP=1.2380 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,941,392B, BPFP=1.2323 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,369,904B, BPFP=0.8695 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,352,752B, BPFP=0.8587 +⌛️ [2/4] FRONTEND: Frontend time: 7.131s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.780s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11488649 21.47604708 + layer.9.1 0.14606862 50.11837321 + layer.19.0 0.08767178 184.57944020 + layer.19.1 0.11443626 133.17746588 + layer.29.0 0.10933029 185.39986188 + layer.29.1 0.10817130 230.34843191 + layer.39.0 52.66717785 2040.65599610 + layer.39.1 62.91127214 2205.99220019 + ------------------------------------------------------------------------------------- + TOTAL 14.53237684 631.46847706 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 12144288 +BPFP 0.9636 bits/point +EBPFP 0.9636 equivalent bits/point +MSE 631.468477 +---------------------- --------------------------------------------------------- +Time: 21.105s Load: 1.194s, Pack+Encode: 7.131s, Decode+Unpack: 12.780s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 631.4685 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00324-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst (31/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.197s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,035,984B, BPFP=0.6576 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,041,372B, BPFP=0.6610 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,732,036B, BPFP=1.0994 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,731,164B, BPFP=1.0989 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,909,116B, BPFP=1.2118 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,907,936B, BPFP=1.2111 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,317,420B, BPFP=0.8362 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,319,408B, BPFP=0.8375 +⌛️ [2/4] FRONTEND: Frontend time: 7.085s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.719s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14509552 87.36321904 + layer.9.1 0.14520687 74.40495511 + layer.19.0 0.12118574 119.82780509 + layer.19.1 0.11709642 151.41403965 + layer.29.0 0.10963326 128.44126787 + layer.29.1 0.10842036 138.81222579 + layer.39.0 53.79489966 2103.28355541 + layer.39.1 62.27410526 2057.88251544 + ------------------------------------------------------------------------------------- + TOTAL 14.60195539 607.67869793 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11994436 +BPFP 0.9517 bits/point +EBPFP 0.9517 equivalent bits/point +MSE 607.678698 +---------------------- --------------------------------------------------------- +Time: 21.001s Load: 1.197s, Pack+Encode: 7.085s, Decode+Unpack: 12.719s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 607.6787 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00325-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst (32/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,179,028B, BPFP=0.7484 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,181,060B, BPFP=0.7497 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,818,676B, BPFP=1.1544 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,816,160B, BPFP=1.1528 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,002,264B, BPFP=1.2709 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,000,076B, BPFP=1.2695 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,347,900B, BPFP=0.8556 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,360,220B, BPFP=0.8634 +⌛️ [2/4] FRONTEND: Frontend time: 7.062s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.766s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14577075 173.09991469 + layer.9.1 0.14541274 260.42606435 + layer.19.0 0.13069581 171.22749431 + layer.19.1 0.13545482 286.47519906 + layer.29.0 0.11331055 117.92663308 + layer.29.1 0.11244963 123.33750406 + layer.39.0 32.27446072 2019.57263568 + layer.39.1 16.59366367 2137.41160221 + ------------------------------------------------------------------------------------- + TOTAL 6.20640234 661.18463093 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 12705384 +BPFP 1.0081 bits/point +EBPFP 1.0081 equivalent bits/point +MSE 661.184631 +---------------------- --------------------------------------------------------- +Time: 21.021s Load: 1.193s, Pack+Encode: 7.062s, Decode+Unpack: 12.766s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 661.1846 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00327-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst (33/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.198s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 987,624B, BPFP=0.6269 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 973,692B, BPFP=0.6181 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,672,240B, BPFP=1.0615 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,664,400B, BPFP=1.0565 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,863,728B, BPFP=1.1830 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,854,836B, BPFP=1.1774 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,249,144B, BPFP=0.7929 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,257,848B, BPFP=0.7984 +⌛️ [2/4] FRONTEND: Frontend time: 7.030s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.628s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14585920 79.02173383 + layer.9.1 0.14576220 111.46959295 + layer.19.0 0.12270736 157.03663268 + layer.19.1 0.12453605 179.35420052 + layer.29.0 0.11393550 152.99082914 + layer.29.1 0.11678154 187.82700276 + layer.39.0 53.83016636 1979.23204420 + layer.39.1 40.65720720 1877.80711732 + ------------------------------------------------------------------------------------- + TOTAL 11.90711942 590.59239417 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11523512 +BPFP 0.9143 bits/point +EBPFP 0.9143 equivalent bits/point +MSE 590.592394 +---------------------- --------------------------------------------------------- +Time: 20.857s Load: 1.198s, Pack+Encode: 7.030s, Decode+Unpack: 12.628s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 590.5924 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00328-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst (34/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.202s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 952,496B, BPFP=0.6046 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 946,068B, BPFP=0.6005 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,724,892B, BPFP=1.0949 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,722,460B, BPFP=1.0933 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,933,848B, BPFP=1.2275 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,936,812B, BPFP=1.2294 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,219,612B, BPFP=0.7741 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,250,076B, BPFP=0.7935 +⌛️ [2/4] FRONTEND: Frontend time: 7.064s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.713s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14568599 37.82265092 + layer.9.1 0.03329684 82.43298566 + layer.19.0 0.11848472 138.17470344 + layer.19.1 0.11973745 114.36687520 + layer.29.0 0.10886538 220.68926714 + layer.29.1 0.10946879 204.84969126 + layer.39.0 14.08931437 1978.29541761 + layer.39.1 9.95616799 1974.54533637 + ------------------------------------------------------------------------------------- + TOTAL 3.08512769 593.89711595 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11686264 +BPFP 0.9272 bits/point +EBPFP 0.9272 equivalent bits/point +MSE 593.897116 +---------------------- --------------------------------------------------------- +Time: 20.980s Load: 1.202s, Pack+Encode: 7.064s, Decode+Unpack: 12.713s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 593.8971 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00332-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst (35/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.191s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 997,744B, BPFP=0.6333 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 992,288B, BPFP=0.6299 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,690,888B, BPFP=1.0733 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,698,292B, BPFP=1.0780 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,880,024B, BPFP=1.1933 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,882,132B, BPFP=1.1947 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,233,128B, BPFP=0.7827 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,238,288B, BPFP=0.7860 +⌛️ [2/4] FRONTEND: Frontend time: 7.111s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.621s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14493624 95.61337138 + layer.9.1 0.14482686 78.78176288 + layer.19.0 0.11946148 206.01344654 + layer.19.1 0.12828579 165.66263812 + layer.29.0 0.10467725 142.15953039 + layer.29.1 0.10613328 168.75072108 + layer.39.0 22.00188902 1891.64998375 + layer.39.1 19.26198661 1823.12902177 + ------------------------------------------------------------------------------------- + TOTAL 5.25152457 571.47005949 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11612784 +BPFP 0.9214 bits/point +EBPFP 0.9214 equivalent bits/point +MSE 571.470059 +---------------------- --------------------------------------------------------- +Time: 20.924s Load: 1.191s, Pack+Encode: 7.111s, Decode+Unpack: 12.621s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 571.4701 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00333-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst (36/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.195s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,001,436B, BPFP=0.6357 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,008,304B, BPFP=0.6400 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,687,840B, BPFP=1.0714 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,686,084B, BPFP=1.0702 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,846,480B, BPFP=1.1721 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,835,020B, BPFP=1.1648 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,211,428B, BPFP=0.7690 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,214,312B, BPFP=0.7708 +⌛️ [2/4] FRONTEND: Frontend time: 7.007s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.413s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14475438 50.56806549 + layer.9.1 0.14492096 83.08735172 + layer.19.0 0.11744098 160.56014381 + layer.19.1 0.11578254 123.94560448 + layer.29.0 0.11402616 193.69105460 + layer.29.1 0.11062706 166.95791355 + layer.39.0 28.92800668 1963.21433214 + layer.39.1 10.80449708 2026.64965876 + ------------------------------------------------------------------------------------- + TOTAL 5.06000698 596.08426557 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11490904 +BPFP 0.9117 bits/point +EBPFP 0.9117 equivalent bits/point +MSE 596.084266 +---------------------- --------------------------------------------------------- +Time: 20.615s Load: 1.195s, Pack+Encode: 7.007s, Decode+Unpack: 12.413s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 596.0843 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00334-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst (37/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.198s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 864,800B, BPFP=0.5489 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 886,156B, BPFP=0.5625 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,551,656B, BPFP=0.9849 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,563,428B, BPFP=0.9924 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,643,180B, BPFP=1.0430 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,662,192B, BPFP=1.0551 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,067,088B, BPFP=0.6773 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,096,316B, BPFP=0.6959 +⌛️ [2/4] FRONTEND: Frontend time: 7.020s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.556s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14546214 17.03003255 + layer.9.1 0.14553630 20.84744170 + layer.19.0 0.04765745 124.64286440 + layer.19.1 0.04191649 100.90099935 + layer.29.0 0.16505912 273.97050699 + layer.29.1 0.15755973 294.52240413 + layer.39.0 42.51041751 1769.43613910 + layer.39.1 31.38856333 1654.36301592 + ------------------------------------------------------------------------------------- + TOTAL 9.32527151 531.96417552 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10334816 +BPFP 0.8200 bits/point +EBPFP 0.8200 equivalent bits/point +MSE 531.964176 +---------------------- --------------------------------------------------------- +Time: 20.774s Load: 1.198s, Pack+Encode: 7.020s, Decode+Unpack: 12.556s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 531.9642 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01347-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst (38/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.207s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 928,632B, BPFP=0.5894 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 923,508B, BPFP=0.5862 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,666,780B, BPFP=1.0580 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,678,460B, BPFP=1.0654 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,811,032B, BPFP=1.1496 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,813,616B, BPFP=1.1512 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,179,936B, BPFP=0.7490 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,185,736B, BPFP=0.7526 +⌛️ [2/4] FRONTEND: Frontend time: 7.161s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.739s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03343166 41.48236411 + layer.9.1 0.03311388 37.20015437 + layer.19.0 0.03842411 54.76655935 + layer.19.1 0.03806642 54.13726337 + layer.29.0 4.26870163 95.61554477 + layer.29.1 4.26552788 91.20897993 + layer.39.0 33.95300821 1798.18573286 + layer.39.1 48.19954501 1660.36886578 + ------------------------------------------------------------------------------------- + TOTAL 11.35372735 479.12068307 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11187700 +BPFP 0.8877 bits/point +EBPFP 0.8877 equivalent bits/point +MSE 479.120683 +---------------------- --------------------------------------------------------- +Time: 21.106s Load: 1.207s, Pack+Encode: 7.161s, Decode+Unpack: 12.739s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 479.1207 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01390-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst (39/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.202s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 950,724B, BPFP=0.6035 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 938,324B, BPFP=0.5956 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,653,324B, BPFP=1.0494 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,634,720B, BPFP=1.0376 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,770,800B, BPFP=1.1240 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,762,320B, BPFP=1.1186 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,169,644B, BPFP=0.7424 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,186,284B, BPFP=0.7530 +⌛️ [2/4] FRONTEND: Frontend time: 7.002s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.511s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14549638 65.78370267 + layer.9.1 0.14520178 73.89880058 + layer.19.0 0.11487435 59.18771835 + layer.19.1 0.11481158 133.62817883 + layer.29.0 0.10827909 92.70365819 + layer.29.1 0.10618535 117.78473351 + layer.39.0 9.83978281 1784.36837829 + layer.39.1 9.67554703 1750.74244394 + ------------------------------------------------------------------------------------- + TOTAL 2.53127230 509.76220179 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11066140 +BPFP 0.8780 bits/point +EBPFP 0.8780 equivalent bits/point +MSE 509.762202 +---------------------- --------------------------------------------------------- +Time: 20.715s Load: 1.202s, Pack+Encode: 7.002s, Decode+Unpack: 12.511s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 509.7622 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01413-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst (40/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.200s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 928,416B, BPFP=0.5893 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 912,864B, BPFP=0.5794 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,657,432B, BPFP=1.0521 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,646,692B, BPFP=1.0452 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,837,764B, BPFP=1.1665 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,827,192B, BPFP=1.1598 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,133,256B, BPFP=0.7193 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,133,752B, BPFP=0.7196 +⌛️ [2/4] FRONTEND: Frontend time: 7.035s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.736s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00092318 29.29202043 + layer.9.1 0.00095285 24.87341313 + layer.19.0 0.08568402 40.46426867 + layer.19.1 0.08404610 59.17564287 + layer.29.0 0.12100375 117.05342054 + layer.29.1 0.12795564 165.72879428 + layer.39.0 12.85620633 1861.92297693 + layer.39.1 12.98640239 1861.25414365 + ------------------------------------------------------------------------------------- + TOTAL 3.28289678 519.97058506 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11077368 +BPFP 0.8789 bits/point +EBPFP 0.8789 equivalent bits/point +MSE 519.970585 +---------------------- --------------------------------------------------------- +Time: 20.970s Load: 1.200s, Pack+Encode: 7.035s, Decode+Unpack: 12.736s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 519.9706 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01421-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst (41/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 947,312B, BPFP=0.6013 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 943,908B, BPFP=0.5991 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,681,228B, BPFP=1.0672 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,678,300B, BPFP=1.0653 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,875,656B, BPFP=1.1906 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,861,748B, BPFP=1.1817 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,208,368B, BPFP=0.7670 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,235,964B, BPFP=0.7845 +⌛️ [2/4] FRONTEND: Frontend time: 7.108s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.687s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00113376 9.05546697 + layer.9.1 0.00100095 21.35089474 + layer.19.0 0.00983371 86.58255809 + layer.19.1 0.00806405 86.16998091 + layer.29.0 4.28365570 77.65125325 + layer.29.1 4.28597952 77.67623192 + layer.39.0 8.41906814 1744.46620084 + layer.39.1 8.59662605 1782.91680208 + ------------------------------------------------------------------------------------- + TOTAL 3.20067024 485.73367360 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11432484 +BPFP 0.9071 bits/point +EBPFP 0.9071 equivalent bits/point +MSE 485.733674 +---------------------- --------------------------------------------------------- +Time: 20.988s Load: 1.193s, Pack+Encode: 7.108s, Decode+Unpack: 12.687s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 485.7337 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01422-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst (42/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.199s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,016,816B, BPFP=0.6454 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,017,072B, BPFP=0.6456 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,723,152B, BPFP=1.0938 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,722,180B, BPFP=1.0932 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,916,360B, BPFP=1.2164 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,918,032B, BPFP=1.2175 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,228,316B, BPFP=0.7797 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,256,532B, BPFP=0.7976 +⌛️ [2/4] FRONTEND: Frontend time: 7.021s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.649s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14550294 63.00428075 + layer.9.1 0.14526658 65.55741185 + layer.19.0 0.11599200 62.96216891 + layer.19.1 0.11361485 44.24527238 + layer.29.0 4.26439454 48.28089962 + layer.29.1 4.25587461 67.55590368 + layer.39.0 8.37236706 1653.62333442 + layer.39.1 8.35116642 1602.31280468 + ------------------------------------------------------------------------------------- + TOTAL 3.22052237 450.94275954 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11798460 +BPFP 0.9361 bits/point +EBPFP 0.9361 equivalent bits/point +MSE 450.942760 +---------------------- --------------------------------------------------------- +Time: 20.868s Load: 1.199s, Pack+Encode: 7.021s, Decode+Unpack: 12.649s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 450.9428 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01441-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst (43/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.191s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 954,164B, BPFP=0.6057 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 952,588B, BPFP=0.6047 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,736,804B, BPFP=1.1024 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,736,780B, BPFP=1.1024 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,917,588B, BPFP=1.2172 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,921,440B, BPFP=1.2196 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,241,888B, BPFP=0.7883 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,236,048B, BPFP=0.7846 +⌛️ [2/4] FRONTEND: Frontend time: 7.140s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.656s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00080404 12.69416995 + layer.9.1 0.00082438 17.20389813 + layer.19.0 0.00843097 62.82156930 + layer.19.1 0.00674472 113.21734847 + layer.29.0 4.27713270 66.48472538 + layer.29.1 4.27133426 63.45764442 + layer.39.0 22.97048921 1639.55801105 + layer.39.1 18.06488920 1619.20750731 + ------------------------------------------------------------------------------------- + TOTAL 6.20008118 449.33060925 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11697300 +BPFP 0.9281 bits/point +EBPFP 0.9281 equivalent bits/point +MSE 449.330609 +---------------------- --------------------------------------------------------- +Time: 20.987s Load: 1.191s, Pack+Encode: 7.140s, Decode+Unpack: 12.656s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 449.3306 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00350-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst (44/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.195s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 938,972B, BPFP=0.5960 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 933,852B, BPFP=0.5928 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,712,828B, BPFP=1.0872 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,721,220B, BPFP=1.0925 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,902,168B, BPFP=1.2074 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,909,588B, BPFP=1.2121 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,250,536B, BPFP=0.7938 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,222,744B, BPFP=0.7761 +⌛️ [2/4] FRONTEND: Frontend time: 7.060s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.647s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14506060 33.24453607 + layer.9.1 0.14523201 49.72819508 + layer.19.0 0.04621643 104.44622400 + layer.19.1 0.04629335 86.46635318 + layer.29.0 4.27940669 102.41166315 + layer.29.1 4.27759670 97.76610741 + layer.39.0 19.91382637 1655.75479363 + layer.39.1 24.01088215 1690.23025674 + ------------------------------------------------------------------------------------- + TOTAL 6.60806429 477.50601616 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11591908 +BPFP 0.9197 bits/point +EBPFP 0.9197 equivalent bits/point +MSE 477.506016 +---------------------- --------------------------------------------------------- +Time: 20.902s Load: 1.195s, Pack+Encode: 7.060s, Decode+Unpack: 12.647s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 477.5060 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00351-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst (45/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.199s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 820,080B, BPFP=0.5205 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 814,316B, BPFP=0.5169 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,449,300B, BPFP=0.9199 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,447,784B, BPFP=0.9190 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,509,104B, BPFP=0.9579 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,519,684B, BPFP=0.9646 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,005,804B, BPFP=0.6384 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,028,624B, BPFP=0.6529 +⌛️ [2/4] FRONTEND: Frontend time: 6.961s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.565s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66985745 16.82000147 + layer.9.1 2.66884121 8.82983502 + layer.19.0 3.21935619 44.29670743 + layer.19.1 3.21606501 40.56604952 + layer.29.0 4.24164606 111.13139828 + layer.29.1 4.23648681 46.50288938 + layer.39.0 8.06392628 1472.66899578 + layer.39.1 8.17747540 1428.34709132 + ------------------------------------------------------------------------------------- + TOTAL 4.56170680 396.14537102 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9594696 +BPFP 0.7613 bits/point +EBPFP 0.7613 equivalent bits/point +MSE 396.145371 +---------------------- --------------------------------------------------------- +Time: 20.725s Load: 1.199s, Pack+Encode: 6.961s, Decode+Unpack: 12.565s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 396.1454 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00358-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst (46/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.198s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 822,896B, BPFP=0.5223 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 829,976B, BPFP=0.5268 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,495,624B, BPFP=0.9493 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,489,912B, BPFP=0.9457 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,640,888B, BPFP=1.0416 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,622,208B, BPFP=1.0297 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,112,748B, BPFP=0.7063 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,104,860B, BPFP=0.7013 +⌛️ [2/4] FRONTEND: Frontend time: 7.012s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.592s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66795352 17.48309407 + layer.9.1 2.66862889 16.97023785 + layer.19.0 3.22250645 54.77308458 + layer.19.1 3.22577319 54.40364499 + layer.29.0 4.25792136 102.00373741 + layer.29.1 4.25014663 83.21275796 + layer.39.0 8.65209937 1644.41111472 + layer.39.1 8.58450170 1704.91290218 + ------------------------------------------------------------------------------------- + TOTAL 4.69119139 459.77132172 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10119112 +BPFP 0.8029 bits/point +EBPFP 0.8029 equivalent bits/point +MSE 459.771322 +---------------------- --------------------------------------------------------- +Time: 20.801s Load: 1.198s, Pack+Encode: 7.012s, Decode+Unpack: 12.592s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 459.7713 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00360-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst (47/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.189s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 961,896B, BPFP=0.6106 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 960,900B, BPFP=0.6099 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,726,160B, BPFP=1.0957 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,730,568B, BPFP=1.0985 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,935,504B, BPFP=1.2286 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,935,052B, BPFP=1.2283 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,198,960B, BPFP=0.7610 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,201,120B, BPFP=0.7624 +⌛️ [2/4] FRONTEND: Frontend time: 7.045s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.675s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00092775 29.24781138 + layer.9.1 0.00093166 24.93307960 + layer.19.0 0.08227225 100.00493581 + layer.19.1 0.08381199 90.87895068 + layer.29.0 0.10725604 51.87450236 + layer.29.1 0.10756977 43.38316034 + layer.39.0 7.96294394 1570.10042249 + layer.39.1 7.95922050 1504.66737082 + ------------------------------------------------------------------------------------- + TOTAL 2.03811674 426.88627919 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11650160 +BPFP 0.9244 bits/point +EBPFP 0.9244 equivalent bits/point +MSE 426.886279 +---------------------- --------------------------------------------------------- +Time: 20.909s Load: 1.189s, Pack+Encode: 7.045s, Decode+Unpack: 12.675s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 426.8863 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00361-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst (48/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.210s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 857,856B, BPFP=0.5445 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 848,032B, BPFP=0.5383 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,561,776B, BPFP=0.9913 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,557,820B, BPFP=0.9888 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,646,852B, BPFP=1.0453 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,631,124B, BPFP=1.0354 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,069,192B, BPFP=0.6787 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,062,468B, BPFP=0.6744 +⌛️ [2/4] FRONTEND: Frontend time: 7.029s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.606s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66308118 8.71104604 + layer.9.1 2.66351027 8.73833711 + layer.19.0 3.21594155 40.29034469 + layer.19.1 3.21498593 40.28930878 + layer.29.0 4.33566519 269.77892428 + layer.29.1 4.34101296 312.73505037 + layer.39.0 8.65310735 1737.42346441 + layer.39.1 8.66575030 1676.06077348 + ------------------------------------------------------------------------------------- + TOTAL 4.71913184 511.75340615 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10235120 +BPFP 0.8121 bits/point +EBPFP 0.8121 equivalent bits/point +MSE 511.753406 +---------------------- --------------------------------------------------------- +Time: 20.845s Load: 1.210s, Pack+Encode: 7.029s, Decode+Unpack: 12.606s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 511.7534 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00384-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst (49/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.194s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 851,772B, BPFP=0.5407 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 852,964B, BPFP=0.5414 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,570,392B, BPFP=0.9968 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,568,400B, BPFP=0.9955 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,638,116B, BPFP=1.0398 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,644,004B, BPFP=1.0435 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 995,684B, BPFP=0.6320 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 977,140B, BPFP=0.6202 +⌛️ [2/4] FRONTEND: Frontend time: 7.142s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.635s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.65959259 8.55363953 + layer.9.1 2.65993726 4.51303586 + layer.19.0 3.20866700 68.64613463 + layer.19.1 3.21007805 39.92160079 + layer.29.0 4.27255361 266.42872522 + layer.29.1 4.27602442 314.00056874 + layer.39.0 19.11658068 1896.59701007 + layer.39.1 9.60360322 1855.44507637 + ------------------------------------------------------------------------------------- + TOTAL 6.12587960 556.76322390 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10098472 +BPFP 0.8013 bits/point +EBPFP 0.8013 equivalent bits/point +MSE 556.763224 +---------------------- --------------------------------------------------------- +Time: 20.971s Load: 1.194s, Pack+Encode: 7.142s, Decode+Unpack: 12.635s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 556.7632 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00385-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst (50/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 884,640B, BPFP=0.5615 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 870,160B, BPFP=0.5523 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,622,444B, BPFP=1.0298 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,615,452B, BPFP=1.0254 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,721,064B, BPFP=1.0924 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,705,256B, BPFP=1.0824 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,086,652B, BPFP=0.6898 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,099,720B, BPFP=0.6980 +⌛️ [2/4] FRONTEND: Frontend time: 6.979s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.588s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67078767 33.67524171 + layer.9.1 2.67131261 16.92750142 + layer.19.0 3.30595795 57.89534246 + layer.19.1 3.30543206 53.66468455 + layer.29.0 0.11228124 177.25684514 + layer.29.1 0.11507649 118.78326089 + layer.39.0 11.41791162 1609.74455639 + layer.39.1 11.38150745 1622.47806305 + ------------------------------------------------------------------------------------- + TOTAL 4.37253339 461.30318695 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10605388 +BPFP 0.8415 bits/point +EBPFP 0.8415 equivalent bits/point +MSE 461.303187 +---------------------- --------------------------------------------------------- +Time: 20.759s Load: 1.193s, Pack+Encode: 6.979s, Decode+Unpack: 12.588s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 461.3032 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00557-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst (51/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.200s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 930,592B, BPFP=0.5907 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 924,280B, BPFP=0.5867 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,614,628B, BPFP=1.0249 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,614,232B, BPFP=1.0246 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,784,024B, BPFP=1.1324 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,780,776B, BPFP=1.1303 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,149,748B, BPFP=0.7298 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,152,096B, BPFP=0.7313 +⌛️ [2/4] FRONTEND: Frontend time: 7.023s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.473s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14463971 45.55385725 + layer.9.1 0.14470460 12.92819330 + layer.19.0 0.12255537 62.74655712 + layer.19.1 0.11825690 117.59336407 + layer.29.0 0.11949990 115.70322148 + layer.29.1 0.11467140 155.19107491 + layer.39.0 10.68243977 1761.20376991 + layer.39.1 10.40156301 1764.36139097 + ------------------------------------------------------------------------------------- + TOTAL 2.73104133 504.41017862 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10950376 +BPFP 0.8688 bits/point +EBPFP 0.8688 equivalent bits/point +MSE 504.410179 +---------------------- --------------------------------------------------------- +Time: 20.695s Load: 1.200s, Pack+Encode: 7.023s, Decode+Unpack: 12.473s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 504.4102 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00001-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst (52/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 975,708B, BPFP=0.6193 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 970,332B, BPFP=0.6159 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,685,428B, BPFP=1.0698 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,675,724B, BPFP=1.0637 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,823,288B, BPFP=1.1573 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,812,992B, BPFP=1.1508 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,190,472B, BPFP=0.7557 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,163,684B, BPFP=0.7386 +⌛️ [2/4] FRONTEND: Frontend time: 7.098s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.784s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14485273 78.63884262 + layer.9.1 0.14484227 62.40604688 + layer.19.0 0.11969613 151.77398846 + layer.19.1 0.11916645 105.28475382 + layer.29.0 0.11480527 210.20677608 + layer.29.1 0.11451660 160.37166883 + layer.39.0 11.00270276 1710.98667533 + layer.39.1 11.01557422 1686.39892753 + ------------------------------------------------------------------------------------- + TOTAL 2.84701956 520.75845994 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11297628 +BPFP 0.8964 bits/point +EBPFP 0.8964 equivalent bits/point +MSE 520.758460 +---------------------- --------------------------------------------------------- +Time: 21.078s Load: 1.196s, Pack+Encode: 7.098s, Decode+Unpack: 12.784s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 520.7585 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00126-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst (53/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.215s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 924,048B, BPFP=0.5865 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 917,636B, BPFP=0.5825 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,571,204B, BPFP=0.9973 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,566,860B, BPFP=0.9946 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,631,424B, BPFP=1.0355 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,632,780B, BPFP=1.0364 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,127,284B, BPFP=0.7155 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,124,400B, BPFP=0.7137 +⌛️ [2/4] FRONTEND: Frontend time: 6.984s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.467s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14481488 62.49677039 + layer.9.1 0.14470567 70.53441359 + layer.19.0 0.03819180 111.43726641 + layer.19.1 0.04002141 78.21791213 + layer.29.0 0.11241068 202.75674358 + layer.29.1 0.11133552 130.04343720 + layer.39.0 31.78807483 1859.18687033 + layer.39.1 43.50691623 1816.71546961 + ------------------------------------------------------------------------------------- + TOTAL 9.48580888 541.42361041 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10495636 +BPFP 0.8328 bits/point +EBPFP 0.8328 equivalent bits/point +MSE 541.423610 +---------------------- --------------------------------------------------------- +Time: 20.666s Load: 1.215s, Pack+Encode: 6.984s, Decode+Unpack: 12.467s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 541.4236 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00127-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst (54/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.197s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 918,780B, BPFP=0.5832 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 925,168B, BPFP=0.5873 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,622,500B, BPFP=1.0299 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,621,044B, BPFP=1.0290 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,759,652B, BPFP=1.1169 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,778,520B, BPFP=1.1289 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,092,444B, BPFP=0.6934 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,128,116B, BPFP=0.7161 +⌛️ [2/4] FRONTEND: Frontend time: 6.979s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.591s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14528272 29.66500955 + layer.9.1 0.14516892 45.92135704 + layer.19.0 0.11319376 127.99240331 + layer.19.1 0.11666145 104.22089292 + layer.29.0 0.21118872 319.90548830 + layer.29.1 0.20646930 378.87743744 + layer.39.0 14.37750853 2116.24780630 + layer.39.1 21.76644002 2176.78436789 + ------------------------------------------------------------------------------------- + TOTAL 4.63523918 662.45184534 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10846224 +BPFP 0.8606 bits/point +EBPFP 0.8606 equivalent bits/point +MSE 662.451845 +---------------------- --------------------------------------------------------- +Time: 20.767s Load: 1.197s, Pack+Encode: 6.979s, Decode+Unpack: 12.591s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 662.4518 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00128-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst (55/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.205s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 890,536B, BPFP=0.5653 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 879,576B, BPFP=0.5583 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,598,204B, BPFP=1.0145 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,590,344B, BPFP=1.0095 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,718,120B, BPFP=1.0906 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,701,736B, BPFP=1.0802 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,190,680B, BPFP=0.7558 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,154,600B, BPFP=0.7329 +⌛️ [2/4] FRONTEND: Frontend time: 7.021s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.671s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14454765 45.74349508 + layer.9.1 0.14475082 45.25794707 + layer.19.0 0.04087094 105.40482816 + layer.19.1 0.11687931 87.21833360 + layer.29.0 0.10817139 103.24803989 + layer.29.1 0.10802081 185.91586773 + layer.39.0 19.80422286 1762.17744556 + layer.39.1 34.29222355 1812.59522262 + ------------------------------------------------------------------------------------- + TOTAL 6.84496092 518.44514747 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10723796 +BPFP 0.8509 bits/point +EBPFP 0.8509 equivalent bits/point +MSE 518.445147 +---------------------- --------------------------------------------------------- +Time: 20.897s Load: 1.205s, Pack+Encode: 7.021s, Decode+Unpack: 12.671s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 518.4451 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00131-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst (56/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.197s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 846,136B, BPFP=0.5371 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 839,464B, BPFP=0.5328 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,547,860B, BPFP=0.9825 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,558,364B, BPFP=0.9892 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,631,660B, BPFP=1.0357 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,646,240B, BPFP=1.0450 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,111,424B, BPFP=0.7055 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,117,308B, BPFP=0.7092 +⌛️ [2/4] FRONTEND: Frontend time: 7.004s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.627s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14462205 33.37242545 + layer.9.1 0.14495783 37.13659815 + layer.19.0 0.04322015 119.73665502 + layer.19.1 0.03788725 92.13662862 + layer.29.0 0.10021623 82.82217866 + layer.29.1 0.10137775 122.22050699 + layer.39.0 58.66958482 1698.70880728 + layer.39.1 72.48303949 1749.16867078 + ------------------------------------------------------------------------------------- + TOTAL 16.46561320 491.91280887 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10298456 +BPFP 0.8171 bits/point +EBPFP 0.8171 equivalent bits/point +MSE 491.912809 +---------------------- --------------------------------------------------------- +Time: 20.827s Load: 1.197s, Pack+Encode: 7.004s, Decode+Unpack: 12.627s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 491.9128 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00132-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst (57/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.195s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 975,584B, BPFP=0.6193 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 983,588B, BPFP=0.6243 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,704,512B, BPFP=1.0819 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,710,444B, BPFP=1.0857 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,845,416B, BPFP=1.1714 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,863,172B, BPFP=1.1826 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,121,892B, BPFP=0.7121 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,143,184B, BPFP=0.7256 +⌛️ [2/4] FRONTEND: Frontend time: 7.024s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.612s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14527927 86.43639300 + layer.9.1 0.14528875 45.96498720 + layer.19.0 0.12591341 123.24964454 + layer.19.1 0.13556211 85.55987975 + layer.29.0 0.11238900 135.60357085 + layer.29.1 0.11028371 125.38525756 + layer.39.0 11.48751193 1649.87292818 + layer.39.1 11.29491489 1682.07848554 + ------------------------------------------------------------------------------------- + TOTAL 2.94464289 491.76889333 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11347792 +BPFP 0.9004 bits/point +EBPFP 0.9004 equivalent bits/point +MSE 491.768893 +---------------------- --------------------------------------------------------- +Time: 20.831s Load: 1.195s, Pack+Encode: 7.024s, Decode+Unpack: 12.612s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 491.7689 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00136-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst (58/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.200s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 958,436B, BPFP=0.6084 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 957,964B, BPFP=0.6081 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,676,496B, BPFP=1.0642 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,675,012B, BPFP=1.0632 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,814,568B, BPFP=1.1518 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,815,188B, BPFP=1.1522 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,162,408B, BPFP=0.7378 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,134,236B, BPFP=0.7200 +⌛️ [2/4] FRONTEND: Frontend time: 7.274s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.512s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14489184 50.32090409 + layer.9.1 0.14511764 66.21241266 + layer.19.0 0.03976490 118.51373091 + layer.19.1 0.11370806 100.82911521 + layer.29.0 0.10933599 80.38379001 + layer.29.1 0.11012027 94.71604851 + layer.39.0 9.10787636 1529.01494963 + layer.39.1 9.00026152 1550.26876828 + ------------------------------------------------------------------------------------- + TOTAL 2.34638457 448.78246491 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11194308 +BPFP 0.8882 bits/point +EBPFP 0.8882 equivalent bits/point +MSE 448.782465 +---------------------- --------------------------------------------------------- +Time: 20.986s Load: 1.200s, Pack+Encode: 7.274s, Decode+Unpack: 12.512s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 448.7825 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00193-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst (59/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.198s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 876,596B, BPFP=0.5564 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 872,764B, BPFP=0.5540 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,605,104B, BPFP=1.0188 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,601,200B, BPFP=1.0164 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,633,832B, BPFP=1.0371 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,631,972B, BPFP=1.0359 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 957,436B, BPFP=0.6077 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 936,540B, BPFP=0.5945 +⌛️ [2/4] FRONTEND: Frontend time: 7.022s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.557s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00083307 17.62149491 + layer.9.1 0.00247171 17.15463266 + layer.19.0 0.00642632 15.69332827 + layer.19.1 0.00641681 39.09294006 + layer.29.0 0.10256791 60.60628758 + layer.29.1 0.10162673 57.19281666 + layer.39.0 8.50517638 1433.21676958 + layer.39.1 8.55767781 1407.50048749 + ------------------------------------------------------------------------------------- + TOTAL 2.16039959 381.00984465 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10115444 +BPFP 0.8026 bits/point +EBPFP 0.8026 equivalent bits/point +MSE 381.009845 +---------------------- --------------------------------------------------------- +Time: 20.776s Load: 1.198s, Pack+Encode: 7.022s, Decode+Unpack: 12.557s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 381.0098 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00196-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst (60/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 894,776B, BPFP=0.5680 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 900,112B, BPFP=0.5713 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,610,488B, BPFP=1.0223 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,607,916B, BPFP=1.0206 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,795,196B, BPFP=1.1395 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,789,080B, BPFP=1.1356 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,141,640B, BPFP=0.7247 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,120,556B, BPFP=0.7113 +⌛️ [2/4] FRONTEND: Frontend time: 6.980s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.604s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00064391 12.60152695 + layer.9.1 0.00065402 12.51018139 + layer.19.0 0.08134466 57.62258287 + layer.19.1 0.08141702 67.37280123 + layer.29.0 0.11551180 173.86007069 + layer.29.1 0.11251285 138.39999391 + layer.39.0 10.61319619 1708.63129672 + layer.39.1 10.43102047 1720.50958726 + ------------------------------------------------------------------------------------- + TOTAL 2.67953762 486.43850513 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10859764 +BPFP 0.8617 bits/point +EBPFP 0.8617 equivalent bits/point +MSE 486.438505 +---------------------- --------------------------------------------------------- +Time: 20.780s Load: 1.196s, Pack+Encode: 6.980s, Decode+Unpack: 12.604s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 486.4385 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00559-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst (61/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.192s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 915,468B, BPFP=0.5811 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 921,896B, BPFP=0.5852 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,661,228B, BPFP=1.0545 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,662,720B, BPFP=1.0554 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,848,360B, BPFP=1.1732 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,858,480B, BPFP=1.1797 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,208,324B, BPFP=0.7670 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,203,184B, BPFP=0.7637 +⌛️ [2/4] FRONTEND: Frontend time: 7.113s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.537s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14441776 42.06532844 + layer.9.1 0.14449203 49.84628900 + layer.19.0 0.11315974 128.35816136 + layer.19.1 0.11435745 76.33119719 + layer.29.0 0.12811458 173.49092054 + layer.29.1 0.12952277 233.09187114 + layer.39.0 31.10682331 1933.50341241 + layer.39.1 16.99297713 1929.68833279 + ------------------------------------------------------------------------------------- + TOTAL 6.10923309 570.79693911 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11279660 +BPFP 0.8950 bits/point +EBPFP 0.8950 equivalent bits/point +MSE 570.796939 +---------------------- --------------------------------------------------------- +Time: 20.842s Load: 1.192s, Pack+Encode: 7.113s, Decode+Unpack: 12.537s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 570.7969 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00570-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst (62/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 907,708B, BPFP=0.5762 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 903,500B, BPFP=0.5735 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,645,392B, BPFP=1.0444 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,646,048B, BPFP=1.0448 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,797,112B, BPFP=1.1407 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,792,372B, BPFP=1.1377 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,114,632B, BPFP=0.7075 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,117,400B, BPFP=0.7093 +⌛️ [2/4] FRONTEND: Frontend time: 7.278s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.822s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00078516 4.61343994 + layer.9.1 0.00079184 4.59476528 + layer.19.0 3.22632161 34.86640803 + layer.19.1 3.22513146 34.52348828 + layer.29.0 0.10494786 84.71005139 + layer.29.1 0.10251782 108.98883856 + layer.39.0 10.88842496 1703.56678583 + layer.39.1 10.78217420 1683.61829704 + ------------------------------------------------------------------------------------- + TOTAL 3.54138686 457.43525929 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10924164 +BPFP 0.8668 bits/point +EBPFP 0.8668 equivalent bits/point +MSE 457.435259 +---------------------- --------------------------------------------------------- +Time: 21.296s Load: 1.196s, Pack+Encode: 7.278s, Decode+Unpack: 12.822s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 457.4353 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00758-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst (63/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.205s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 878,660B, BPFP=0.5577 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 876,960B, BPFP=0.5567 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,517,572B, BPFP=0.9633 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,517,080B, BPFP=0.9630 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,670,336B, BPFP=1.0602 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,662,568B, BPFP=1.0553 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,066,168B, BPFP=0.6767 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,069,512B, BPFP=0.6789 +⌛️ [2/4] FRONTEND: Frontend time: 6.984s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.600s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14561636 42.30081553 + layer.9.1 0.14552785 58.65538674 + layer.19.0 0.04069186 44.95979749 + layer.19.1 0.03840616 63.60776527 + layer.29.0 0.11346353 119.03394134 + layer.29.1 0.11182956 127.94311627 + layer.39.0 10.19697364 1805.94995125 + layer.39.1 10.11578978 1860.80532987 + ------------------------------------------------------------------------------------- + TOTAL 2.61353734 515.40701297 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10258856 +BPFP 0.8140 bits/point +EBPFP 0.8140 equivalent bits/point +MSE 515.407013 +---------------------- --------------------------------------------------------- +Time: 20.789s Load: 1.205s, Pack+Encode: 6.984s, Decode+Unpack: 12.600s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 515.4070 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00764-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst (64/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.206s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 913,756B, BPFP=0.5800 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 916,208B, BPFP=0.5816 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,662,984B, BPFP=1.0556 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,654,320B, BPFP=1.0501 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,818,432B, BPFP=1.1542 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,808,852B, BPFP=1.1482 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,146,988B, BPFP=0.7281 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,135,316B, BPFP=0.7206 +⌛️ [2/4] FRONTEND: Frontend time: 7.007s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.565s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14399445 24.92533819 + layer.9.1 0.14558028 37.51703668 + layer.19.0 0.03837104 80.92687683 + layer.19.1 0.04376782 85.92443939 + layer.29.0 0.11695251 123.84416640 + layer.29.1 0.13128335 173.25024374 + layer.39.0 11.28613757 1666.61147221 + layer.39.1 11.84408769 1601.50861228 + ------------------------------------------------------------------------------------- + TOTAL 2.96877184 474.31352322 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11056856 +BPFP 0.8773 bits/point +EBPFP 0.8773 equivalent bits/point +MSE 474.313523 +---------------------- --------------------------------------------------------- +Time: 20.777s Load: 1.206s, Pack+Encode: 7.007s, Decode+Unpack: 12.565s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 474.3135 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00845-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst (65/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.199s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 904,180B, BPFP=0.5739 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 901,044B, BPFP=0.5719 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,704,436B, BPFP=1.0819 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,700,148B, BPFP=1.0792 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,939,352B, BPFP=1.2310 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,923,924B, BPFP=1.2212 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,238,248B, BPFP=0.7860 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,296,176B, BPFP=0.8227 +⌛️ [2/4] FRONTEND: Frontend time: 7.290s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.752s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14502686 20.83713083 + layer.9.1 0.03259508 17.08291482 + layer.19.0 0.11326540 90.52914771 + layer.19.1 0.11324834 76.24566847 + layer.29.0 0.12250664 201.26836204 + layer.29.1 0.12058897 170.77413065 + layer.39.0 16.17915050 2123.50178746 + layer.39.1 21.66230805 2108.13422164 + ------------------------------------------------------------------------------------- + TOTAL 4.81108623 601.04667045 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11607508 +BPFP 0.9210 bits/point +EBPFP 0.9210 equivalent bits/point +MSE 601.046670 +---------------------- --------------------------------------------------------- +Time: 21.240s Load: 1.199s, Pack+Encode: 7.290s, Decode+Unpack: 12.752s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 601.0467 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00870-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst (66/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.199s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 885,276B, BPFP=0.5619 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 881,696B, BPFP=0.5597 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,606,024B, BPFP=1.0194 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,609,668B, BPFP=1.0217 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,756,216B, BPFP=1.1148 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,766,140B, BPFP=1.1211 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,162,864B, BPFP=0.7381 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,130,740B, BPFP=0.7177 +⌛️ [2/4] FRONTEND: Frontend time: 7.027s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.579s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66733019 16.65711072 + layer.9.1 2.66763138 4.79016284 + layer.19.0 3.22293078 21.35065354 + layer.19.1 3.22376992 16.21934666 + layer.29.0 4.27658332 53.58704196 + layer.29.1 4.27160529 58.86788572 + layer.39.0 7.81683598 1552.11748456 + layer.39.1 9.86231960 1558.68297043 + ------------------------------------------------------------------------------------- + TOTAL 4.75112581 410.28408206 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10798624 +BPFP 0.8568 bits/point +EBPFP 0.8568 equivalent bits/point +MSE 410.284082 +---------------------- --------------------------------------------------------- +Time: 20.805s Load: 1.199s, Pack+Encode: 7.027s, Decode+Unpack: 12.579s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 410.2841 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00153-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst (67/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.197s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 865,664B, BPFP=0.5495 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 866,252B, BPFP=0.5499 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,524,668B, BPFP=0.9678 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,537,452B, BPFP=0.9759 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,610,632B, BPFP=1.0223 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,627,820B, BPFP=1.0333 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,064,036B, BPFP=0.6754 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,083,688B, BPFP=0.6879 +⌛️ [2/4] FRONTEND: Frontend time: 6.999s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.438s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14549763 29.77868053 + layer.9.1 0.14520254 49.85257556 + layer.19.0 0.04746155 68.58230419 + layer.19.1 0.04383140 77.42693269 + layer.29.0 4.26247378 87.06379997 + layer.29.1 4.25497898 103.31794361 + layer.39.0 7.94138086 1556.78095548 + layer.39.1 7.86439079 1496.97058824 + ------------------------------------------------------------------------------------- + TOTAL 3.08815219 433.72172253 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10180212 +BPFP 0.8077 bits/point +EBPFP 0.8077 equivalent bits/point +MSE 433.721723 +---------------------- --------------------------------------------------------- +Time: 20.634s Load: 1.197s, Pack+Encode: 6.999s, Decode+Unpack: 12.438s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 433.7217 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00167-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst (68/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.197s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 838,560B, BPFP=0.5323 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 831,240B, BPFP=0.5276 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,482,600B, BPFP=0.9411 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,472,744B, BPFP=0.9348 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,563,968B, BPFP=0.9927 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,553,736B, BPFP=0.9862 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,002,044B, BPFP=0.6360 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,018,072B, BPFP=0.6462 +⌛️ [2/4] FRONTEND: Frontend time: 6.957s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.443s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11283708 25.53767113 + layer.9.1 0.11300174 24.70076627 + layer.19.0 3.22718329 39.59124147 + layer.19.1 3.22892155 58.83380728 + layer.29.0 4.26448309 119.79353876 + layer.29.1 4.25758082 78.07062480 + layer.39.0 9.82393946 1848.76974326 + layer.39.1 9.78394007 1754.82499188 + ------------------------------------------------------------------------------------- + TOTAL 4.35148589 493.76529810 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 9762964 +BPFP 0.7746 bits/point +EBPFP 0.7746 equivalent bits/point +MSE 493.765298 +---------------------- --------------------------------------------------------- +Time: 20.597s Load: 1.197s, Pack+Encode: 6.957s, Decode+Unpack: 12.443s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 493.7653 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00208-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst (69/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.186s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,043,640B, BPFP=0.6625 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,032,372B, BPFP=0.6553 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,665,856B, BPFP=1.0574 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,665,120B, BPFP=1.0569 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,769,284B, BPFP=1.1231 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,761,228B, BPFP=1.1179 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,171,080B, BPFP=0.7433 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,186,332B, BPFP=0.7530 +⌛️ [2/4] FRONTEND: Frontend time: 7.229s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.494s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14503861 131.56317030 + layer.9.1 0.14483112 136.39685570 + layer.19.0 0.11529889 87.92999472 + layer.19.1 0.11517203 161.50199058 + layer.29.0 0.11961639 95.91727941 + layer.29.1 0.11795276 178.69012025 + layer.39.0 83.84633978 2102.36707832 + layer.39.1 174.87768118 2092.81117972 + ------------------------------------------------------------------------------------- + TOTAL 32.43524135 623.39720862 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11294912 +BPFP 0.8962 bits/point +EBPFP 0.8962 equivalent bits/point +MSE 623.397209 +---------------------- --------------------------------------------------------- +Time: 20.910s Load: 1.186s, Pack+Encode: 7.229s, Decode+Unpack: 12.494s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 623.3972 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01201-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst (70/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 845,928B, BPFP=0.5370 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 837,092B, BPFP=0.5313 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,548,452B, BPFP=0.9829 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,544,280B, BPFP=0.9802 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,699,564B, BPFP=1.0788 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,689,344B, BPFP=1.0723 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,155,776B, BPFP=0.7336 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,178,372B, BPFP=0.7480 +⌛️ [2/4] FRONTEND: Frontend time: 7.017s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.586s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.81453913 33.71865098 + layer.9.1 0.14528001 45.21631256 + layer.19.0 3.26598681 40.07763243 + layer.19.1 0.04116655 86.63977697 + layer.29.0 4.28557138 222.98789405 + layer.29.1 4.28198282 198.13166233 + layer.39.0 74.89367180 1763.42785180 + layer.39.1 42.04871577 1803.33604160 + ------------------------------------------------------------------------------------- + TOTAL 16.47211428 524.19197784 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10498808 +BPFP 0.8330 bits/point +EBPFP 0.8330 equivalent bits/point +MSE 524.191978 +---------------------- --------------------------------------------------------- +Time: 20.799s Load: 1.196s, Pack+Encode: 7.017s, Decode+Unpack: 12.586s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 524.1920 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01296-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst (71/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.197s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 863,664B, BPFP=0.5482 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 872,992B, BPFP=0.5541 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,590,312B, BPFP=1.0095 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,602,544B, BPFP=1.0172 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,719,204B, BPFP=1.0913 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,730,476B, BPFP=1.0984 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,078,916B, BPFP=0.6848 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,080,956B, BPFP=0.6861 +⌛️ [2/4] FRONTEND: Frontend time: 6.985s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.727s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66710422 13.08059672 + layer.9.1 2.66812426 12.89363244 + layer.19.0 3.22059776 58.66215063 + layer.19.1 3.22546153 82.66906179 + layer.29.0 0.11226317 199.94436545 + layer.29.1 0.11257672 163.90940851 + layer.39.0 59.39237691 1793.43451414 + layer.39.1 37.52358222 1800.52811180 + ------------------------------------------------------------------------------------- + TOTAL 13.61526085 515.64023019 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10539064 +BPFP 0.8362 bits/point +EBPFP 0.8362 equivalent bits/point +MSE 515.640230 +---------------------- --------------------------------------------------------- +Time: 20.909s Load: 1.197s, Pack+Encode: 6.985s, Decode+Unpack: 12.727s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 515.6402 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01313-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst (72/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.192s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 849,764B, BPFP=0.5394 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 852,340B, BPFP=0.5410 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,623,520B, BPFP=1.0305 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,627,692B, BPFP=1.0332 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,789,592B, BPFP=1.1359 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,802,380B, BPFP=1.1441 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,113,304B, BPFP=0.7067 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,100,132B, BPFP=0.6983 +⌛️ [2/4] FRONTEND: Frontend time: 6.985s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.549s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14519123 4.53382740 + layer.9.1 0.14511500 4.51328658 + layer.19.0 0.03974548 17.06257744 + layer.19.1 0.03981401 11.67705329 + layer.29.0 4.26343511 95.91069833 + layer.29.1 4.25610090 110.96121425 + layer.39.0 7.90972018 1634.81036724 + layer.39.1 8.05601540 1658.16363341 + ------------------------------------------------------------------------------------- + TOTAL 3.10689216 442.20408224 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10758724 +BPFP 0.8536 bits/point +EBPFP 0.8536 equivalent bits/point +MSE 442.204082 +---------------------- --------------------------------------------------------- +Time: 20.726s Load: 1.192s, Pack+Encode: 6.985s, Decode+Unpack: 12.549s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 442.2041 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00032-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst (73/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.201s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 865,864B, BPFP=0.5496 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 859,456B, BPFP=0.5455 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,614,180B, BPFP=1.0246 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,602,504B, BPFP=1.0172 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,805,844B, BPFP=1.1463 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,793,880B, BPFP=1.1387 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,113,556B, BPFP=0.7068 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,094,008B, BPFP=0.6944 +⌛️ [2/4] FRONTEND: Frontend time: 7.014s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.459s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14586543 25.09724112 + layer.9.1 0.14572574 53.86837321 + layer.19.0 0.03953905 44.46780549 + layer.19.1 0.03760033 53.80093232 + layer.29.0 0.10448607 79.93324972 + layer.29.1 0.10697372 99.76143565 + layer.39.0 14.19073468 1831.64088398 + layer.39.1 8.92149669 1812.73090673 + ------------------------------------------------------------------------------------- + TOTAL 2.96155271 500.16260353 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10749292 +BPFP 0.8529 bits/point +EBPFP 0.8529 equivalent bits/point +MSE 500.162604 +---------------------- --------------------------------------------------------- +Time: 20.673s Load: 1.201s, Pack+Encode: 7.014s, Decode+Unpack: 12.459s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 500.1626 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00041-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst (74/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.194s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 906,768B, BPFP=0.5756 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 911,404B, BPFP=0.5785 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,672,900B, BPFP=1.0619 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,671,916B, BPFP=1.0612 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,831,488B, BPFP=1.1625 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,828,804B, BPFP=1.1608 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,154,276B, BPFP=0.7327 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,163,280B, BPFP=0.7384 +⌛️ [2/4] FRONTEND: Frontend time: 7.124s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.720s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00251883 28.90246181 + layer.9.1 0.14409062 25.14433702 + layer.19.0 0.12740102 151.63360213 + layer.19.1 0.12254588 156.28006175 + layer.29.0 4.25147928 126.10047327 + layer.29.1 4.25065697 101.86807361 + layer.39.0 9.21805114 1622.32580435 + layer.39.1 9.03214690 1649.99561261 + ------------------------------------------------------------------------------------- + TOTAL 3.39361133 482.78130332 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11140836 +BPFP 0.8840 bits/point +EBPFP 0.8840 equivalent bits/point +MSE 482.781303 +---------------------- --------------------------------------------------------- +Time: 21.037s Load: 1.194s, Pack+Encode: 7.124s, Decode+Unpack: 12.720s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 482.7813 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00042-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst (75/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.200s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 998,080B, BPFP=0.6335 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 995,136B, BPFP=0.6317 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,707,112B, BPFP=1.0836 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,703,928B, BPFP=1.0816 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,874,572B, BPFP=1.1899 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,879,944B, BPFP=1.1933 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,279,936B, BPFP=0.8124 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,275,096B, BPFP=0.8094 +⌛️ [2/4] FRONTEND: Frontend time: 7.093s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.700s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14550905 92.39101601 + layer.9.1 0.14590163 99.79135522 + layer.19.0 0.12839093 142.90942883 + layer.19.1 0.12422524 119.99850707 + layer.29.0 0.11695262 226.53355541 + layer.29.1 0.11389293 185.27285099 + layer.39.0 10.18180439 1761.52453689 + layer.39.1 10.42432323 1837.48895028 + ------------------------------------------------------------------------------------- + TOTAL 2.67262500 558.23877509 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11713804 +BPFP 0.9294 bits/point +EBPFP 0.9294 equivalent bits/point +MSE 558.238775 +---------------------- --------------------------------------------------------- +Time: 20.993s Load: 1.200s, Pack+Encode: 7.093s, Decode+Unpack: 12.700s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 558.2388 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00620-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst (76/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 912,836B, BPFP=0.5794 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 927,400B, BPFP=0.5887 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,660,712B, BPFP=1.0541 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,669,108B, BPFP=1.0595 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,871,000B, BPFP=1.1876 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,878,636B, BPFP=1.1925 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,184,068B, BPFP=0.7516 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,203,564B, BPFP=0.7640 +⌛️ [2/4] FRONTEND: Frontend time: 7.035s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.687s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14509775 46.00747481 + layer.9.1 0.14508723 38.27670519 + layer.19.0 0.11633494 100.00811464 + layer.19.1 0.11804005 124.19185692 + layer.29.0 0.15409572 241.07639340 + layer.29.1 0.14997486 227.55506581 + layer.39.0 9.23291952 1817.26698083 + layer.39.1 9.22304726 1830.53201170 + ------------------------------------------------------------------------------------- + TOTAL 2.41057467 553.11432541 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11307324 +BPFP 0.8972 bits/point +EBPFP 0.8972 equivalent bits/point +MSE 553.114325 +---------------------- --------------------------------------------------------- +Time: 20.914s Load: 1.193s, Pack+Encode: 7.035s, Decode+Unpack: 12.687s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 553.1143 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00636-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst (77/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.197s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 962,216B, BPFP=0.6108 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 963,352B, BPFP=0.6115 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,681,596B, BPFP=1.0674 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,686,148B, BPFP=1.0703 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,873,504B, BPFP=1.1892 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,878,692B, BPFP=1.1925 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,228,132B, BPFP=0.7796 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,227,232B, BPFP=0.7790 +⌛️ [2/4] FRONTEND: Frontend time: 7.177s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.635s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14497060 65.78225544 + layer.9.1 0.14492971 82.86395028 + layer.19.0 0.11929473 137.48821904 + layer.19.1 0.11869117 133.03096563 + layer.29.0 0.13715227 276.57759181 + layer.29.1 0.14278979 241.37717338 + layer.39.0 9.99110525 1929.08986025 + layer.39.1 10.01170034 1976.68167046 + ------------------------------------------------------------------------------------- + TOTAL 2.60132923 605.36146079 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11500872 +BPFP 0.9125 bits/point +EBPFP 0.9125 equivalent bits/point +MSE 605.361461 +---------------------- --------------------------------------------------------- +Time: 21.009s Load: 1.197s, Pack+Encode: 7.177s, Decode+Unpack: 12.635s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 605.3615 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00637-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst (78/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.194s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 901,572B, BPFP=0.5723 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 898,976B, BPFP=0.5706 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,643,260B, BPFP=1.0431 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,642,992B, BPFP=1.0429 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,844,188B, BPFP=1.1706 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,856,832B, BPFP=1.1786 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,248,400B, BPFP=0.7924 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,247,352B, BPFP=0.7918 +⌛️ [2/4] FRONTEND: Frontend time: 7.008s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.610s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14589856 25.36358974 + layer.9.1 0.03321603 62.42160587 + layer.19.0 0.11866178 150.99040258 + layer.19.1 0.11267978 178.88879184 + layer.29.0 0.10803594 106.40756012 + layer.29.1 0.10714094 140.39445889 + layer.39.0 11.58943751 1773.29476763 + layer.39.1 9.70079103 1776.09002275 + ------------------------------------------------------------------------------------- + TOTAL 2.73948269 526.73139993 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11283572 +BPFP 0.8953 bits/point +EBPFP 0.8953 equivalent bits/point +MSE 526.731400 +---------------------- --------------------------------------------------------- +Time: 20.811s Load: 1.194s, Pack+Encode: 7.008s, Decode+Unpack: 12.610s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 526.7314 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00410-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst (79/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.202s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 850,376B, BPFP=0.5398 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 857,340B, BPFP=0.5442 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,625,420B, BPFP=1.0317 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,628,624B, BPFP=1.0338 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,793,692B, BPFP=1.1385 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,801,660B, BPFP=1.1436 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,154,960B, BPFP=0.7331 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,119,592B, BPFP=0.7107 +⌛️ [2/4] FRONTEND: Frontend time: 6.994s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.556s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03308301 29.18489753 + layer.9.1 0.14566304 20.89561286 + layer.19.0 0.03810260 66.81377966 + layer.19.1 0.03780774 53.88246466 + layer.29.0 0.11592613 143.09547652 + layer.29.1 0.11717217 153.97216241 + layer.39.0 9.98032847 1702.00584985 + layer.39.1 9.70849498 1697.67387065 + ------------------------------------------------------------------------------------- + TOTAL 2.52207227 483.44051427 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10831664 +BPFP 0.8594 bits/point +EBPFP 0.8594 equivalent bits/point +MSE 483.440514 +---------------------- --------------------------------------------------------- +Time: 20.752s Load: 1.202s, Pack+Encode: 6.994s, Decode+Unpack: 12.556s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 483.4405 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00411-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst (80/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.197s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 830,700B, BPFP=0.5273 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 822,940B, BPFP=0.5224 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,590,040B, BPFP=1.0093 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,583,776B, BPFP=1.0053 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,678,252B, BPFP=1.0653 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,675,960B, BPFP=1.0638 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 999,796B, BPFP=0.6346 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,005,560B, BPFP=0.6383 +⌛️ [2/4] FRONTEND: Frontend time: 6.997s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.492s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14555504 9.01407685 + layer.9.1 0.14557384 20.69454699 + layer.19.0 0.03995539 94.72369597 + layer.19.1 0.04542811 113.11652990 + layer.29.0 0.12033866 122.92048871 + layer.29.1 0.13252172 141.84843191 + layer.39.0 10.37566776 1656.68004550 + layer.39.1 9.84188447 1637.43288918 + ------------------------------------------------------------------------------------- + TOTAL 2.60586562 474.55383813 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10187024 +BPFP 0.8083 bits/point +EBPFP 0.8083 equivalent bits/point +MSE 474.553838 +---------------------- --------------------------------------------------------- +Time: 20.686s Load: 1.197s, Pack+Encode: 6.997s, Decode+Unpack: 12.492s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 474.5538 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00412-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst (81/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 949,840B, BPFP=0.6029 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 958,740B, BPFP=0.6086 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,702,316B, BPFP=1.0805 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,701,416B, BPFP=1.0800 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,901,848B, BPFP=1.2072 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,905,044B, BPFP=1.2092 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,308,376B, BPFP=0.8305 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,304,668B, BPFP=0.8281 +⌛️ [2/4] FRONTEND: Frontend time: 7.056s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.671s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14470203 66.04309697 + layer.9.1 0.14481130 53.81494760 + layer.19.0 0.11257574 86.97884506 + layer.19.1 0.11422884 96.30726966 + layer.29.0 0.10456927 141.33639706 + layer.29.1 0.10551051 165.17028559 + layer.39.0 10.36536069 1901.64965876 + layer.39.1 11.81531702 1830.30744231 + ------------------------------------------------------------------------------------- + TOTAL 2.86338443 542.70099288 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11732248 +BPFP 0.9309 bits/point +EBPFP 0.9309 equivalent bits/point +MSE 542.700993 +---------------------- --------------------------------------------------------- +Time: 20.920s Load: 1.193s, Pack+Encode: 7.056s, Decode+Unpack: 12.671s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 542.7010 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00413-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst (82/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.195s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 943,452B, BPFP=0.5989 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 942,512B, BPFP=0.5983 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,682,500B, BPFP=1.0680 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,677,840B, BPFP=1.0650 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,777,828B, BPFP=1.1285 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,770,932B, BPFP=1.1241 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,117,276B, BPFP=0.7092 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,111,208B, BPFP=0.7053 +⌛️ [2/4] FRONTEND: Frontend time: 7.338s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.668s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14530003 37.69799622 + layer.9.1 0.14546206 41.57526609 + layer.19.0 0.11891763 87.37998659 + layer.19.1 0.11677460 81.62314145 + layer.29.0 4.29725807 180.04838317 + layer.29.1 4.29692800 117.63272871 + layer.39.0 11.61914761 1714.50341241 + layer.39.1 11.22064282 1689.65859604 + ------------------------------------------------------------------------------------- + TOTAL 3.99505385 493.76493884 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11023548 +BPFP 0.8746 bits/point +EBPFP 0.8746 equivalent bits/point +MSE 493.764939 +---------------------- --------------------------------------------------------- +Time: 21.201s Load: 1.195s, Pack+Encode: 7.338s, Decode+Unpack: 12.668s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 493.7649 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00433-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst (83/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 835,052B, BPFP=0.5300 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 826,504B, BPFP=0.5246 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,536,572B, BPFP=0.9753 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,540,180B, BPFP=0.9776 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,656,240B, BPFP=1.0513 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,662,876B, BPFP=1.0555 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,148,920B, BPFP=0.7293 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,116,836B, BPFP=0.7089 +⌛️ [2/4] FRONTEND: Frontend time: 6.982s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.515s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67106792 4.58236545 + layer.9.1 2.67195307 16.80495409 + layer.19.0 0.08237472 49.82401182 + layer.19.1 0.08192194 77.45802019 + layer.29.0 0.11152953 175.24356110 + layer.29.1 0.11703055 243.87390315 + layer.39.0 163.01811830 1816.50146246 + layer.39.1 58.15221299 1854.75073123 + ------------------------------------------------------------------------------------- + TOTAL 28.36327613 529.87987619 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10323180 +BPFP 0.8191 bits/point +EBPFP 0.8191 equivalent bits/point +MSE 529.879876 +---------------------- --------------------------------------------------------- +Time: 20.690s Load: 1.193s, Pack+Encode: 6.982s, Decode+Unpack: 12.515s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 529.8799 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00443-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst (84/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.211s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 945,768B, BPFP=0.6003 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 948,384B, BPFP=0.6020 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,618,552B, BPFP=1.0274 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,623,104B, BPFP=1.0303 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,762,228B, BPFP=1.1186 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,762,936B, BPFP=1.1190 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,146,512B, BPFP=0.7277 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,107,104B, BPFP=0.7027 +⌛️ [2/4] FRONTEND: Frontend time: 7.018s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.585s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14620710 87.53853185 + layer.9.1 0.14642976 75.07432158 + layer.19.0 0.11726453 127.88699423 + layer.19.1 0.11958517 137.60716607 + layer.29.0 0.10693079 101.86814470 + layer.29.1 0.10826971 116.36126909 + layer.39.0 43.01306569 1912.48310042 + layer.39.1 17.12450997 1909.90591485 + ------------------------------------------------------------------------------------- + TOTAL 7.61028284 558.59068035 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10914588 +BPFP 0.8660 bits/point +EBPFP 0.8660 equivalent bits/point +MSE 558.590680 +---------------------- --------------------------------------------------------- +Time: 20.814s Load: 1.211s, Pack+Encode: 7.018s, Decode+Unpack: 12.585s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 558.5907 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00444-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst (85/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.199s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 837,352B, BPFP=0.5315 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 826,536B, BPFP=0.5246 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,539,600B, BPFP=0.9773 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,539,528B, BPFP=0.9772 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,677,184B, BPFP=1.0646 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,667,380B, BPFP=1.0584 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,061,896B, BPFP=0.6740 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,086,040B, BPFP=0.6894 +⌛️ [2/4] FRONTEND: Frontend time: 6.969s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.530s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03173930 37.13112660 + layer.9.1 0.03345565 20.79892270 + layer.19.0 3.26068347 59.96732816 + layer.19.1 3.26087326 81.64632759 + layer.29.0 4.24610771 71.01536094 + layer.29.1 4.24089229 76.13230216 + layer.39.0 8.81319124 1575.63015925 + layer.39.1 8.71779153 1504.39665258 + ------------------------------------------------------------------------------------- + TOTAL 4.07559181 428.33977250 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10235516 +BPFP 0.8121 bits/point +EBPFP 0.8121 equivalent bits/point +MSE 428.339772 +---------------------- --------------------------------------------------------- +Time: 20.698s Load: 1.199s, Pack+Encode: 6.969s, Decode+Unpack: 12.530s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 428.3398 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00445-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst (86/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.200s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 878,144B, BPFP=0.5574 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 879,088B, BPFP=0.5580 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,641,344B, BPFP=1.0418 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,645,108B, BPFP=1.0442 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,846,960B, BPFP=1.1724 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,861,444B, BPFP=1.1816 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,191,388B, BPFP=0.7562 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,204,056B, BPFP=0.7643 +⌛️ [2/4] FRONTEND: Frontend time: 6.987s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.624s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00078599 16.77707081 + layer.9.1 0.00079117 20.56974630 + layer.19.0 0.00795310 85.96157987 + layer.19.1 0.00811505 58.00375772 + layer.29.0 4.25797468 102.60344898 + layer.29.1 4.25504309 94.13256622 + layer.39.0 81.06806549 1731.49301267 + layer.39.1 44.82015254 1698.64429639 + ------------------------------------------------------------------------------------- + TOTAL 16.80236014 476.02318487 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11147532 +BPFP 0.8845 bits/point +EBPFP 0.8845 equivalent bits/point +MSE 476.023185 +---------------------- --------------------------------------------------------- +Time: 20.811s Load: 1.200s, Pack+Encode: 6.987s, Decode+Unpack: 12.624s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 476.0232 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00446-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst (87/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.199s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 887,240B, BPFP=0.5632 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 881,684B, BPFP=0.5596 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,663,240B, BPFP=1.0557 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,657,096B, BPFP=1.0518 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,872,076B, BPFP=1.1883 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,873,396B, BPFP=1.1891 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,218,720B, BPFP=0.7736 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,193,696B, BPFP=0.7577 +⌛️ [2/4] FRONTEND: Frontend time: 7.121s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.972s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00082351 21.10111310 + layer.9.1 0.02968625 33.25805878 + layer.19.0 0.00841222 58.28758734 + layer.19.1 0.03743129 76.72788532 + layer.29.0 4.28408194 56.65359421 + layer.29.1 4.28564945 125.34868581 + layer.39.0 8.35370986 1604.93922652 + layer.39.1 8.52557915 1643.54306142 + ------------------------------------------------------------------------------------- + TOTAL 3.19067171 452.48240156 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11247148 +BPFP 0.8924 bits/point +EBPFP 0.8924 equivalent bits/point +MSE 452.482402 +---------------------- --------------------------------------------------------- +Time: 21.292s Load: 1.199s, Pack+Encode: 7.121s, Decode+Unpack: 12.972s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 452.4824 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00447-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst (88/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.218s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 886,652B, BPFP=0.5628 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 890,136B, BPFP=0.5650 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,611,432B, BPFP=1.0229 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,621,688B, BPFP=1.0294 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,766,020B, BPFP=1.1210 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,771,608B, BPFP=1.1245 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,109,216B, BPFP=0.7041 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,126,604B, BPFP=0.7151 +⌛️ [2/4] FRONTEND: Frontend time: 7.017s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.651s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14529990 29.33334772 + layer.9.1 0.14524076 20.79363270 + layer.19.0 0.03780325 109.20118013 + layer.19.1 0.03783790 87.14158474 + layer.29.0 4.32098184 219.28325073 + layer.29.1 4.32100596 185.90041030 + layer.39.0 9.32673680 2015.56044849 + layer.39.1 9.31823369 1990.07491063 + ------------------------------------------------------------------------------------- + TOTAL 3.45664251 582.16109568 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10783356 +BPFP 0.8556 bits/point +EBPFP 0.8556 equivalent bits/point +MSE 582.161096 +---------------------- --------------------------------------------------------- +Time: 20.886s Load: 1.218s, Pack+Encode: 7.017s, Decode+Unpack: 12.651s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 582.1611 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00461-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst (89/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.201s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 914,028B, BPFP=0.5802 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 907,112B, BPFP=0.5758 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,633,952B, BPFP=1.0372 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,627,728B, BPFP=1.0332 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,810,036B, BPFP=1.1489 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,797,656B, BPFP=1.1411 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,151,432B, BPFP=0.7309 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,155,412B, BPFP=0.7334 +⌛️ [2/4] FRONTEND: Frontend time: 7.005s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.606s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14497286 33.68101032 + layer.9.1 0.14497296 50.39258003 + layer.19.0 0.03962668 179.93012675 + layer.19.1 0.11751332 169.58529006 + layer.29.0 0.14529291 223.26681833 + layer.29.1 0.16241527 194.64053867 + layer.39.0 11.40179406 1979.44848879 + layer.39.1 13.03458244 1926.12463438 + ------------------------------------------------------------------------------------- + TOTAL 3.14889631 594.63368592 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10997356 +BPFP 0.8726 bits/point +EBPFP 0.8726 equivalent bits/point +MSE 594.633686 +---------------------- --------------------------------------------------------- +Time: 20.812s Load: 1.201s, Pack+Encode: 7.005s, Decode+Unpack: 12.606s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 594.6337 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00462-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst (90/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.199s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 977,708B, BPFP=0.6206 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 983,820B, BPFP=0.6245 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,702,096B, BPFP=1.0804 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,702,356B, BPFP=1.0806 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,881,476B, BPFP=1.1943 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,881,456B, BPFP=1.1943 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,212,344B, BPFP=0.7695 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,205,396B, BPFP=0.7651 +⌛️ [2/4] FRONTEND: Frontend time: 7.015s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.499s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14513344 16.78027122 + layer.9.1 0.03283094 21.11282423 + layer.19.0 0.11544709 110.27918833 + layer.19.1 0.11326018 197.01645271 + layer.29.0 0.14483232 208.07604810 + layer.29.1 0.14672551 223.20874634 + layer.39.0 10.02784076 1816.54533637 + layer.39.1 15.62606130 1883.26178096 + ------------------------------------------------------------------------------------- + TOTAL 3.29401644 559.53508103 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11546652 +BPFP 0.9162 bits/point +EBPFP 0.9162 equivalent bits/point +MSE 559.535081 +---------------------- --------------------------------------------------------- +Time: 20.713s Load: 1.199s, Pack+Encode: 7.015s, Decode+Unpack: 12.499s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 559.5351 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00463-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst (91/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.191s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,013,252B, BPFP=0.6432 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,013,432B, BPFP=0.6433 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,742,000B, BPFP=1.1057 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,741,612B, BPFP=1.1055 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,908,036B, BPFP=1.2111 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,916,804B, BPFP=1.2167 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,237,432B, BPFP=0.7855 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,236,388B, BPFP=0.7848 +⌛️ [2/4] FRONTEND: Frontend time: 7.031s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.678s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14492303 37.33326140 + layer.9.1 0.14484742 45.55252681 + layer.19.0 0.11740684 160.40841323 + layer.19.1 0.11489933 174.17535343 + layer.29.0 0.12072669 186.60651609 + layer.29.1 0.12118037 202.13479038 + layer.39.0 10.74778980 1840.94296393 + layer.39.1 11.83662176 1945.11358466 + ------------------------------------------------------------------------------------- + TOTAL 2.91854940 574.03342624 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11808956 +BPFP 0.9370 bits/point +EBPFP 0.9370 equivalent bits/point +MSE 574.033426 +---------------------- --------------------------------------------------------- +Time: 20.900s Load: 1.191s, Pack+Encode: 7.031s, Decode+Unpack: 12.678s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 574.0334 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00464-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst (92/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.198s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,056,872B, BPFP=0.6708 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,052,660B, BPFP=0.6682 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,759,408B, BPFP=1.1168 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,753,452B, BPFP=1.1130 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,937,008B, BPFP=1.2295 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,921,164B, BPFP=1.2195 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,255,528B, BPFP=0.7969 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,245,564B, BPFP=0.7906 +⌛️ [2/4] FRONTEND: Frontend time: 7.062s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.679s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14465000 37.77610345 + layer.9.1 0.14489275 70.33322839 + layer.19.0 0.11978787 221.86033474 + layer.19.1 0.12819003 286.92214413 + layer.29.0 0.12519148 251.54960188 + layer.29.1 0.13018718 147.74205801 + layer.39.0 10.77894586 1962.20685733 + layer.39.1 10.25834823 2011.66135847 + ------------------------------------------------------------------------------------- + TOTAL 2.72877418 623.75646080 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11981656 +BPFP 0.9507 bits/point +EBPFP 0.9507 equivalent bits/point +MSE 623.756461 +---------------------- --------------------------------------------------------- +Time: 20.938s Load: 1.198s, Pack+Encode: 7.062s, Decode+Unpack: 12.679s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 623.7565 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00465-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst (93/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.192s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 891,740B, BPFP=0.5660 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 903,768B, BPFP=0.5737 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,603,100B, BPFP=1.0176 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,617,044B, BPFP=1.0264 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,743,932B, BPFP=1.1070 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,755,340B, BPFP=1.1142 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,145,672B, BPFP=0.7272 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,188,388B, BPFP=0.7543 +⌛️ [2/4] FRONTEND: Frontend time: 6.982s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.389s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14553812 24.98577399 + layer.9.1 0.14559401 21.13266270 + layer.19.0 0.04492324 67.97208624 + layer.19.1 0.04213941 123.09829989 + layer.29.0 4.25320263 87.19582792 + layer.29.1 4.25391672 117.42791274 + layer.39.0 8.72311137 1539.45482613 + layer.39.1 8.87262096 1676.32304192 + ------------------------------------------------------------------------------------- + TOTAL 3.31013081 457.19880394 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 10848984 +BPFP 0.8608 bits/point +EBPFP 0.8608 equivalent bits/point +MSE 457.198804 +---------------------- --------------------------------------------------------- +Time: 20.563s Load: 1.192s, Pack+Encode: 6.982s, Decode+Unpack: 12.389s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 457.1988 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00471-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst (94/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.197s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 946,304B, BPFP=0.6007 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 946,684B, BPFP=0.6009 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,650,600B, BPFP=1.0477 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,649,552B, BPFP=1.0471 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,830,892B, BPFP=1.1622 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,829,908B, BPFP=1.1615 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,189,164B, BPFP=0.7548 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,196,600B, BPFP=0.7595 +⌛️ [2/4] FRONTEND: Frontend time: 7.024s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.616s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14518122 54.11138081 + layer.9.1 0.14529820 45.77777157 + layer.19.0 0.11833418 123.79054274 + layer.19.1 0.12038008 170.37246100 + layer.29.0 4.31360161 280.68067517 + layer.29.1 4.31792870 281.29974407 + layer.39.0 9.40764201 1781.88836529 + layer.39.1 11.30764416 1781.60139747 + ------------------------------------------------------------------------------------- + TOTAL 3.73450127 564.94029226 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11239704 +BPFP 0.8918 bits/point +EBPFP 0.8918 equivalent bits/point +MSE 564.940292 +---------------------- --------------------------------------------------------- +Time: 20.837s Load: 1.197s, Pack+Encode: 7.024s, Decode+Unpack: 12.616s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 564.9403 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00472-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst (95/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.204s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,052,368B, BPFP=0.6680 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,060,536B, BPFP=0.6732 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,749,056B, BPFP=1.1102 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,749,660B, BPFP=1.1106 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,921,044B, BPFP=1.2194 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,933,364B, BPFP=1.2272 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,204,544B, BPFP=0.7646 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,203,976B, BPFP=0.7642 +⌛️ [2/4] FRONTEND: Frontend time: 7.296s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.681s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00315138 107.56741550 + layer.9.1 0.00505826 140.72682402 + layer.19.0 0.09147678 142.40076576 + layer.19.1 0.09143778 161.19800130 + layer.29.0 0.11015094 126.94253737 + layer.29.1 0.11338039 200.76015600 + layer.39.0 9.14784464 1635.77396815 + layer.39.1 8.98944348 1601.94735132 + ------------------------------------------------------------------------------------- + TOTAL 2.31899296 514.66462743 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11874548 +BPFP 0.9422 bits/point +EBPFP 0.9422 equivalent bits/point +MSE 514.664627 +---------------------- --------------------------------------------------------- +Time: 21.182s Load: 1.204s, Pack+Encode: 7.296s, Decode+Unpack: 12.681s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 514.6646 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00475-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst (96/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.203s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,090,132B, BPFP=0.6920 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,081,944B, BPFP=0.6868 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,791,332B, BPFP=1.1370 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,785,644B, BPFP=1.1334 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,002,672B, BPFP=1.2712 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,987,496B, BPFP=1.2616 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,299,744B, BPFP=0.8250 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,273,296B, BPFP=0.8082 +⌛️ [2/4] FRONTEND: Frontend time: 7.044s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.632s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14591913 63.62066339 + layer.9.1 0.03347605 111.21865860 + layer.19.0 0.12173996 193.83835717 + layer.19.1 0.12099332 173.76641209 + layer.29.0 0.11078974 123.38229201 + layer.29.1 0.11776269 152.14135115 + layer.39.0 10.17800795 1855.68280793 + layer.39.1 9.88744998 1777.58986025 + ------------------------------------------------------------------------------------- + TOTAL 2.58951735 556.40505032 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 12312260 +BPFP 0.9769 bits/point +EBPFP 0.9769 equivalent bits/point +MSE 556.405050 +---------------------- --------------------------------------------------------- +Time: 20.879s Load: 1.203s, Pack+Encode: 7.044s, Decode+Unpack: 12.632s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 556.4051 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00476-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst (97/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.200s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 879,380B, BPFP=0.5582 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 881,760B, BPFP=0.5597 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,648,048B, BPFP=1.0461 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,652,088B, BPFP=1.0487 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,875,888B, BPFP=1.1907 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,883,192B, BPFP=1.1954 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,111,632B, BPFP=0.7056 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,129,276B, BPFP=0.7168 +⌛️ [2/4] FRONTEND: Frontend time: 6.971s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.647s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66405266 8.76409652 + layer.9.1 2.66543197 8.60289230 + layer.19.0 3.22131407 53.72916497 + layer.19.1 3.22426883 43.99832934 + layer.29.0 4.27224607 108.19116631 + layer.29.1 4.27784520 84.22183742 + layer.39.0 8.94937744 1631.19922002 + layer.39.1 8.82170070 1696.06467338 + ------------------------------------------------------------------------------------- + TOTAL 4.76202962 454.34642253 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11061264 +BPFP 0.8776 bits/point +EBPFP 0.8776 equivalent bits/point +MSE 454.346423 +---------------------- --------------------------------------------------------- +Time: 20.818s Load: 1.200s, Pack+Encode: 6.971s, Decode+Unpack: 12.647s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 454.3464 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00643-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst (98/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.196s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 946,576B, BPFP=0.6008 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 947,100B, BPFP=0.6012 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,692,972B, BPFP=1.0746 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,691,644B, BPFP=1.0738 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,908,948B, BPFP=1.2117 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,895,708B, BPFP=1.2033 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,206,020B, BPFP=0.7655 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,221,604B, BPFP=0.7754 +⌛️ [2/4] FRONTEND: Frontend time: 7.018s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.665s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00104424 21.24984512 + layer.9.1 0.00091568 41.79047164 + layer.19.0 0.08171424 76.50574829 + layer.19.1 0.08373584 62.77571498 + layer.29.0 4.26071267 57.11046169 + layer.29.1 4.26438533 53.08297449 + layer.39.0 8.39843369 1584.43061423 + layer.39.1 8.51949380 1589.42703932 + ------------------------------------------------------------------------------------- + TOTAL 3.20130444 435.79660872 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11510572 +BPFP 0.9133 bits/point +EBPFP 0.9133 equivalent bits/point +MSE 435.796609 +---------------------- --------------------------------------------------------- +Time: 20.879s Load: 1.196s, Pack+Encode: 7.018s, Decode+Unpack: 12.665s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 435.7966 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00644-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst (99/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.204s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,014,464B, BPFP=0.6439 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,009,912B, BPFP=0.6410 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,726,360B, BPFP=1.0958 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,732,468B, BPFP=1.0997 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,929,636B, BPFP=1.2248 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,930,440B, BPFP=1.2253 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,240,716B, BPFP=0.7875 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,236,652B, BPFP=0.7850 +⌛️ [2/4] FRONTEND: Frontend time: 7.275s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.716s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14622840 86.87540624 + layer.9.1 0.03344178 65.95851276 + layer.19.0 0.12675888 173.67582060 + layer.19.1 0.12382618 207.59148521 + layer.29.0 0.12223263 200.68664690 + layer.29.1 0.12797405 160.97545296 + layer.39.0 10.69978368 1838.04907377 + layer.39.1 8.63538768 1780.09522262 + ------------------------------------------------------------------------------------- + TOTAL 2.50195416 564.23845263 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11820648 +BPFP 0.9379 bits/point +EBPFP 0.9379 equivalent bits/point +MSE 564.238453 +---------------------- --------------------------------------------------------- +Time: 21.195s Load: 1.204s, Pack+Encode: 7.275s, Decode+Unpack: 12.716s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 564.2385 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00272-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst (100/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.200s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 989,060B, BPFP=0.6278 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 991,444B, BPFP=0.6293 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 1,662,000B, BPFP=1.0550 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 1,664,472B, BPFP=1.0565 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 1,801,752B, BPFP=1.1437 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 1,803,656B, BPFP=1.1449 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,161,980B, BPFP=0.7376 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,149,984B, BPFP=0.7300 +⌛️ [2/4] FRONTEND: Frontend time: 7.023s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.875s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14492994 110.81849204 + layer.9.1 0.14498602 74.61428035 + layer.19.0 0.12957112 120.19513731 + layer.19.1 0.13054295 133.58179639 + layer.29.0 0.16610158 349.16253656 + layer.29.1 0.14872770 305.88265762 + layer.39.0 16.52878844 1892.67435814 + layer.39.1 24.55764797 1838.49057524 + ------------------------------------------------------------------------------------- + TOTAL 5.24391197 603.17747921 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 11224348 +BPFP 0.8906 bits/point +EBPFP 0.8906 equivalent bits/point +MSE 603.177479 +---------------------- --------------------------------------------------------- +Time: 21.097s Load: 1.200s, Pack+Encode: 7.023s, Decode+Unpack: 12.875s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 603.1775 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst + to output-fixed/dinov3-total/lambda0.01/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00278-wflip.zst +------------------------ ---------------------------- +TOTAL PROCESSING SUMMARY +------------------------ ---------------------------- +Total files 100 +Avg BPFP 0.8792 bits/point +Avg EBPFP 0.8792 equivalent bits/point +Avg MSE 522.778215 +Avg Time 20.889s +------------------------ ---------------------------- diff --git a/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log b/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log new file mode 100644 index 0000000000000000000000000000000000000000..2270ea815a9aefd3fde979c07a1b0eea38c7de8f --- /dev/null +++ b/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log @@ -0,0 +1,15744 @@ +Experiment: dtufc_hyperprior-featurecoding_dinov3-total_individual +Log file: output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dtufc_hyperprior-featurecoding_dinov3-total_individual.log +DTUFCCodecConfig: + arch: hyperprior-featurecoding + handler: dinov3-total + checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.02_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar + transform_type: kmeans + transform_mapping:featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json + bit_depth: 8 + device: cuda:0 +Loading checkpoint: codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.02_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar +Checkpoint epoch: 598 +Loaded hyperprior-featurecoding (1-channel) on cuda:0 +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_9_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.9' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_9_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_19_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.19' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_19_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_29_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.29' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_29_0.json +Loaded quantization points from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_39_0.json: torch.Size([256]) +Loaded per-key quantization points for key 'layer.39' from featurecoding_utils/transform_mapping/kmeans10samples-8bits/dinov3_total/dep_fewshot-8bit_layer_39_0.json +Loaded per-key mappings: model=dinov3-total + Keys: ['layer.9', 'layer.19', 'layer.29', 'layer.39'] +---------------- ----------------------------------------------------------------------------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +Checkpoint codec_weights/hyperprior_hybrid/bmshj2018-hyperprior_lambda0.02_epochs600_lr0.0001_bs180_patch256-256_checkpoint_best.pth.tar +Transform type kmeans +Transform config featurecoding_utils/transform_mapping/kmeans10samples-8bits/mapping.json +Input ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features +Output output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test +---------------- ----------------------------------------------------------------------------------------------------------------------------- +Files found: 100 +---------------------------------------------------------------------- + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst (1/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.239s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,629,368B, BPFP=1.0342 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,639,056B, BPFP=1.0404 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,029,660B, BPFP=1.2883 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,038,344B, BPFP=1.2938 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,296,068B, BPFP=1.4574 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,302,488B, BPFP=1.4615 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,562,960B, BPFP=0.9921 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,569,680B, BPFP=0.9964 +⌛️ [2/4] FRONTEND: Frontend time: 8.088s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.418s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14561056 12.48219146 + layer.9.1 0.14522085 4.45286285 + layer.19.0 3.25142184 19.94484786 + layer.19.1 3.25206135 6.04669727 + layer.29.0 4.23946030 33.36554731 + layer.29.1 4.24539299 47.02443025 + layer.39.0 32.17105490 1190.77266818 + layer.39.1 19.15684032 1211.07166071 + ------------------------------------------------------------------------------------- + TOTAL 8.32588289 315.64511324 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15067624 +BPFP 1.1955 bits/point +EBPFP 1.1955 equivalent bits/point +MSE 315.645113 +---------------------- --------------------------------------------------------- +Time: 22.745s Load: 1.239s, Pack+Encode: 8.088s, Decode+Unpack: 13.418s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 315.6451 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00512-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00512-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst (2/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.253s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,743,520B, BPFP=1.1067 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,742,852B, BPFP=1.1063 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,282,616B, BPFP=1.4489 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,275,228B, BPFP=1.4442 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,583,784B, BPFP=1.6401 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,569,132B, BPFP=1.6308 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,752,252B, BPFP=1.1122 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,736,160B, BPFP=1.1020 +⌛️ [2/4] FRONTEND: Frontend time: 7.634s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.811s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03307773 4.33159264 + layer.9.1 0.03291117 4.44576476 + layer.19.0 0.04156009 57.06595812 + layer.19.1 0.03760627 43.07174196 + layer.29.0 4.28582750 37.64844512 + layer.29.1 4.28551552 38.62557127 + layer.39.0 9.83402183 1083.62235944 + layer.39.1 9.85397836 1146.29614885 + ------------------------------------------------------------------------------------- + TOTAL 3.55056231 301.88844777 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16685544 +BPFP 1.3239 bits/point +EBPFP 1.3239 equivalent bits/point +MSE 301.888448 +---------------------- --------------------------------------------------------- +Time: 21.697s Load: 1.253s, Pack+Encode: 7.634s, Decode+Unpack: 12.811s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 301.8884 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00657-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00657-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst (3/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.260s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,778,008B, BPFP=1.1286 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,769,160B, BPFP=1.1230 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,287,440B, BPFP=1.4520 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,292,960B, BPFP=1.4555 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,610,368B, BPFP=1.6569 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,613,656B, BPFP=1.6590 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,704,540B, BPFP=1.0820 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,712,060B, BPFP=1.0867 +⌛️ [2/4] FRONTEND: Frontend time: 7.884s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.336s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00083900 4.35956669 + layer.9.1 0.00259629 8.56535130 + layer.19.0 0.00955961 24.48874462 + layer.19.1 0.08538111 28.92095588 + layer.29.0 0.11631418 43.30204948 + layer.29.1 0.11200302 68.69285221 + layer.39.0 14.47657393 1281.63544036 + layer.39.1 13.08093694 1282.50999350 + ------------------------------------------------------------------------------------- + TOTAL 3.48552551 342.80936926 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16768192 +BPFP 1.3305 bits/point +EBPFP 1.3305 equivalent bits/point +MSE 342.809369 +---------------------- --------------------------------------------------------- +Time: 22.481s Load: 1.260s, Pack+Encode: 7.884s, Decode+Unpack: 13.336s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 342.8094 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00676-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00676-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst (4/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.243s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,579,556B, BPFP=1.0026 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,594,344B, BPFP=1.0120 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,125,404B, BPFP=1.3491 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,135,552B, BPFP=1.3555 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,334,296B, BPFP=1.4817 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,330,348B, BPFP=1.4792 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,569,468B, BPFP=0.9962 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,595,624B, BPFP=1.0128 +⌛️ [2/4] FRONTEND: Frontend time: 7.949s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.408s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.70169554 4.35211791 + layer.9.1 0.03294074 4.33220676 + layer.19.0 3.25671692 29.43432625 + layer.19.1 3.25834093 15.90580694 + layer.29.0 0.10810242 158.84806630 + layer.29.1 0.10661203 95.57179274 + layer.39.0 8.95005916 1078.47627559 + layer.39.1 8.98756017 1079.50666233 + ------------------------------------------------------------------------------------- + TOTAL 3.42525349 308.30340685 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15264592 +BPFP 1.2111 bits/point +EBPFP 1.2111 equivalent bits/point +MSE 308.303407 +---------------------- --------------------------------------------------------- +Time: 22.601s Load: 1.243s, Pack+Encode: 7.949s, Decode+Unpack: 13.408s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 308.3034 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00688-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00688-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst (5/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.257s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,695,964B, BPFP=1.0765 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,672,508B, BPFP=1.0616 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,220,732B, BPFP=1.4096 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,231,296B, BPFP=1.4163 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,515,096B, BPFP=1.5965 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,524,808B, BPFP=1.6026 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,636,976B, BPFP=1.0391 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,665,832B, BPFP=1.0574 +⌛️ [2/4] FRONTEND: Frontend time: 7.696s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.669s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14527870 4.32178004 + layer.9.1 0.14521496 4.31321663 + layer.19.0 0.03964342 24.45375975 + layer.19.1 0.03956446 15.27506627 + layer.29.0 0.12258449 97.19315689 + layer.29.1 0.12735008 174.03199139 + layer.39.0 32.94776263 1308.76893078 + layer.39.1 29.25669534 1262.05833604 + ------------------------------------------------------------------------------------- + TOTAL 7.85301176 361.30202972 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16163212 +BPFP 1.2824 bits/point +EBPFP 1.2824 equivalent bits/point +MSE 361.302030 +---------------------- --------------------------------------------------------- +Time: 21.623s Load: 1.257s, Pack+Encode: 7.696s, Decode+Unpack: 12.669s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 361.3020 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00693-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00693-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst (6/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.230s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,631,244B, BPFP=1.0354 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,645,328B, BPFP=1.0444 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,117,872B, BPFP=1.3443 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,087,524B, BPFP=1.3251 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,318,048B, BPFP=1.4714 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,316,524B, BPFP=1.4704 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,552,704B, BPFP=0.9856 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,564,340B, BPFP=0.9930 +⌛️ [2/4] FRONTEND: Frontend time: 7.584s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.334s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66816044 4.39482355 + layer.9.1 2.66817504 12.97201896 + layer.19.0 3.22262959 20.26228622 + layer.19.1 3.22037432 15.27072839 + layer.29.0 4.30448692 146.94029290 + layer.29.1 4.31085282 118.26202470 + layer.39.0 38.33931691 1232.48115047 + layer.39.1 57.25219370 1266.15502112 + ------------------------------------------------------------------------------------- + TOTAL 14.49827372 352.09229329 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15233584 +BPFP 1.2087 bits/point +EBPFP 1.2087 equivalent bits/point +MSE 352.092293 +---------------------- --------------------------------------------------------- +Time: 22.148s Load: 1.230s, Pack+Encode: 7.584s, Decode+Unpack: 13.334s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 352.0923 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bathroom-rgb_00731-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bathroom-rgb_00731-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst (7/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.246s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,736,552B, BPFP=1.1023 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,741,708B, BPFP=1.1055 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,186,344B, BPFP=1.3878 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,181,644B, BPFP=1.3848 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,429,520B, BPFP=1.5421 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,430,056B, BPFP=1.5425 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,587,460B, BPFP=1.0076 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,591,904B, BPFP=1.0105 +⌛️ [2/4] FRONTEND: Frontend time: 7.802s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.784s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11332920 4.40156523 + layer.9.1 0.00092169 4.39644629 + layer.19.0 3.23006092 20.00081883 + layer.19.1 3.23257961 24.84318888 + layer.29.0 4.28548854 48.42475422 + layer.29.1 4.27808990 23.66764757 + layer.39.0 10.57841825 1126.95376991 + layer.39.1 20.33118703 1177.93906402 + ------------------------------------------------------------------------------------- + TOTAL 5.75625939 303.82840687 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15885188 +BPFP 1.2604 bits/point +EBPFP 1.2604 equivalent bits/point +MSE 303.828407 +---------------------- --------------------------------------------------------- +Time: 21.832s Load: 1.246s, Pack+Encode: 7.802s, Decode+Unpack: 12.784s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 303.8284 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00061-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00061-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst (8/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.243s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,817,104B, BPFP=1.1534 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,822,796B, BPFP=1.1570 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,205,308B, BPFP=1.3998 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,211,236B, BPFP=1.4036 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,382,844B, BPFP=1.5125 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,395,532B, BPFP=1.5206 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,569,496B, BPFP=0.9962 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,572,920B, BPFP=0.9984 +⌛️ [2/4] FRONTEND: Frontend time: 7.833s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.814s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14443954 12.56587179 + layer.9.1 0.14435121 16.80882226 + layer.19.0 0.03807715 38.98950378 + layer.19.1 0.03781311 20.33393677 + layer.29.0 0.10781899 74.81515071 + layer.29.1 0.10618912 55.79180208 + layer.39.0 9.30898666 1292.53973026 + layer.39.1 9.83625107 1274.18069548 + ------------------------------------------------------------------------------------- + TOTAL 2.46549086 348.25318914 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15977236 +BPFP 1.2677 bits/point +EBPFP 1.2677 equivalent bits/point +MSE 348.253189 +---------------------- --------------------------------------------------------- +Time: 21.890s Load: 1.243s, Pack+Encode: 7.833s, Decode+Unpack: 12.814s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 348.2532 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00175-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00175-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst (9/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.247s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,798,708B, BPFP=1.1417 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,801,180B, BPFP=1.1433 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,306,656B, BPFP=1.4641 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,302,296B, BPFP=1.4614 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,601,328B, BPFP=1.6512 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,600,896B, BPFP=1.6509 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,725,660B, BPFP=1.0954 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,704,816B, BPFP=1.0821 +⌛️ [2/4] FRONTEND: Frontend time: 7.854s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.715s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14543928 8.62637487 + layer.9.1 0.14562574 4.47640223 + layer.19.0 0.11552505 38.63803522 + layer.19.1 0.12052174 33.95764442 + layer.29.0 0.10841144 39.88687998 + layer.29.1 0.10845811 45.16149049 + layer.39.0 9.17501701 1131.27153071 + layer.39.1 9.20635778 1196.83937277 + ------------------------------------------------------------------------------------- + TOTAL 2.39066952 312.35721634 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16841540 +BPFP 1.3363 bits/point +EBPFP 1.3363 equivalent bits/point +MSE 312.357216 +---------------------- --------------------------------------------------------- +Time: 21.816s Load: 1.247s, Pack+Encode: 7.854s, Decode+Unpack: 12.715s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 312.3572 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00186-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00186-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst (10/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.267s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,710,136B, BPFP=1.0855 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,720,944B, BPFP=1.0924 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,047,532B, BPFP=1.2997 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,035,924B, BPFP=1.2923 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,187,756B, BPFP=1.3887 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,202,056B, BPFP=1.3978 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,472,912B, BPFP=0.9349 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,507,420B, BPFP=0.9568 +⌛️ [2/4] FRONTEND: Frontend time: 7.758s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.772s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.78226592 8.65381573 + layer.9.1 2.78427046 8.52438899 + layer.19.0 3.22580366 19.74489915 + layer.19.1 3.22969594 15.33766402 + layer.29.0 4.29525448 83.25202104 + layer.29.1 0.11349234 77.44032844 + layer.39.0 8.89338553 1132.53542411 + layer.39.1 8.88767087 1182.05516737 + ------------------------------------------------------------------------------------- + TOTAL 4.27647990 315.94296361 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 14884680 +BPFP 1.1810 bits/point +EBPFP 1.1810 equivalent bits/point +MSE 315.942964 +---------------------- --------------------------------------------------------- +Time: 21.797s Load: 1.267s, Pack+Encode: 7.758s, Decode+Unpack: 12.772s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 315.9430 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00282-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00282-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst (11/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.246s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,687,320B, BPFP=1.0710 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,681,156B, BPFP=1.0671 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,088,000B, BPFP=1.3254 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,077,832B, BPFP=1.3189 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,264,092B, BPFP=1.4371 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,258,880B, BPFP=1.4338 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,473,256B, BPFP=0.9351 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,495,872B, BPFP=0.9495 +⌛️ [2/4] FRONTEND: Frontend time: 7.642s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.851s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14514284 12.74630576 + layer.9.1 0.14518188 8.87198875 + layer.19.0 0.04057091 33.72911927 + layer.19.1 0.04041447 19.99399527 + layer.29.0 4.25641542 38.49149181 + layer.29.1 4.26613502 53.57114783 + layer.39.0 12.58558458 1290.82417940 + layer.39.1 8.96866240 1332.05695483 + ------------------------------------------------------------------------------------- + TOTAL 3.80601344 348.78564786 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15026408 +BPFP 1.1923 bits/point +EBPFP 1.1923 equivalent bits/point +MSE 348.785648 +---------------------- --------------------------------------------------------- +Time: 21.738s Load: 1.246s, Pack+Encode: 7.642s, Decode+Unpack: 12.851s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 348.7856 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00524-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00524-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst (12/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.269s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,705,772B, BPFP=1.0827 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,685,604B, BPFP=1.0699 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,147,044B, BPFP=1.3628 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,146,704B, BPFP=1.3626 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,397,808B, BPFP=1.5220 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,402,276B, BPFP=1.5248 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,548,160B, BPFP=0.9827 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,580,784B, BPFP=1.0034 +⌛️ [2/4] FRONTEND: Frontend time: 7.618s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.696s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67015388 4.39027747 + layer.9.1 0.00076871 4.36657751 + layer.19.0 3.22151687 5.84823133 + layer.19.1 3.22388957 15.03891143 + layer.29.0 4.24084786 28.82350656 + layer.29.1 4.24602234 23.90042300 + layer.39.0 7.87160790 1006.58994150 + layer.39.1 9.85764150 1012.61724082 + ------------------------------------------------------------------------------------- + TOTAL 4.41655608 262.69688870 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15614152 +BPFP 1.2389 bits/point +EBPFP 1.2389 equivalent bits/point +MSE 262.696889 +---------------------- --------------------------------------------------------- +Time: 21.583s Load: 1.269s, Pack+Encode: 7.618s, Decode+Unpack: 12.696s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 262.6969 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00538-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00538-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst (13/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.173s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,801,928B, BPFP=1.1438 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,806,588B, BPFP=1.1467 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,275,608B, BPFP=1.4444 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,273,060B, BPFP=1.4428 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,561,220B, BPFP=1.6257 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,552,916B, BPFP=1.6205 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,718,684B, BPFP=1.0909 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,716,760B, BPFP=1.0897 +⌛️ [2/4] FRONTEND: Frontend time: 7.759s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.379s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00068237 4.49911548 + layer.9.1 0.00070576 4.49697383 + layer.19.0 0.00823322 34.89332396 + layer.19.1 0.08594799 34.06333025 + layer.29.0 0.12200666 170.78926308 + layer.29.1 0.12451052 136.76409652 + layer.39.0 55.99513528 1532.51169971 + layer.39.1 28.81185256 1509.06792330 + ------------------------------------------------------------------------------------- + TOTAL 10.64363429 428.38571577 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16706764 +BPFP 1.3256 bits/point +EBPFP 1.3256 equivalent bits/point +MSE 428.385716 +---------------------- --------------------------------------------------------- +Time: 22.311s Load: 1.173s, Pack+Encode: 7.759s, Decode+Unpack: 13.379s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 428.3857 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00927-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00927-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst (14/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.244s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,748,260B, BPFP=1.1097 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,764,920B, BPFP=1.1203 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,142,096B, BPFP=1.3597 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,156,412B, BPFP=1.3688 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,316,468B, BPFP=1.4704 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,355,732B, BPFP=1.4953 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,565,628B, BPFP=0.9938 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,591,992B, BPFP=1.0105 +⌛️ [2/4] FRONTEND: Frontend time: 7.808s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.810s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14556937 8.76341607 + layer.9.1 0.03327741 8.54826320 + layer.19.0 0.11590617 52.78381947 + layer.19.1 0.11733878 43.49115921 + layer.29.0 0.11334742 54.45250548 + layer.29.1 4.29039579 68.25510339 + layer.39.0 9.10722066 1202.54582385 + layer.39.1 44.52401893 1227.91607085 + ------------------------------------------------------------------------------------- + TOTAL 7.30588432 333.34452019 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15641508 +BPFP 1.2411 bits/point +EBPFP 1.2411 equivalent bits/point +MSE 333.344520 +---------------------- --------------------------------------------------------- +Time: 21.862s Load: 1.244s, Pack+Encode: 7.808s, Decode+Unpack: 12.810s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 333.3445 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_00933-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_00933-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst (15/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.241s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,873,784B, BPFP=1.1894 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,871,620B, BPFP=1.1880 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,275,564B, BPFP=1.4444 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,287,640B, BPFP=1.4521 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,473,284B, BPFP=1.5699 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,478,664B, BPFP=1.5733 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,503,896B, BPFP=0.9546 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,529,556B, BPFP=0.9709 +⌛️ [2/4] FRONTEND: Frontend time: 7.810s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.069s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11318631 4.78588367 + layer.9.1 0.11319129 24.89709691 + layer.19.0 0.00665199 15.26917452 + layer.19.1 0.00853768 15.04369363 + layer.29.0 4.27225940 19.62448077 + layer.29.1 4.27324961 34.43136324 + layer.39.0 14.80262837 1209.46766331 + layer.39.1 16.56649765 1328.42419565 + ------------------------------------------------------------------------------------- + TOTAL 5.01952529 331.49294396 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16294008 +BPFP 1.2928 bits/point +EBPFP 1.2928 equivalent bits/point +MSE 331.492944 +---------------------- --------------------------------------------------------- +Time: 22.121s Load: 1.241s, Pack+Encode: 7.810s, Decode+Unpack: 13.069s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 331.4929 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01095-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01095-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst (16/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.232s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,773,424B, BPFP=1.1257 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,767,500B, BPFP=1.1219 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,199,432B, BPFP=1.3961 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,187,392B, BPFP=1.3884 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,414,528B, BPFP=1.5326 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,417,572B, BPFP=1.5346 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,586,976B, BPFP=1.0073 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,578,416B, BPFP=1.0019 +⌛️ [2/4] FRONTEND: Frontend time: 7.712s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.807s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00065117 4.37052977 + layer.9.1 0.00066201 4.37089729 + layer.19.0 0.00984582 15.34820721 + layer.19.1 0.01156107 6.67625922 + layer.29.0 4.26547583 39.34902350 + layer.29.1 4.26296603 43.81420113 + layer.39.0 11.21169412 1260.24260643 + layer.39.1 9.31977106 1268.76689958 + ------------------------------------------------------------------------------------- + TOTAL 3.63532839 330.36732802 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15925240 +BPFP 1.2636 bits/point +EBPFP 1.2636 equivalent bits/point +MSE 330.367328 +---------------------- --------------------------------------------------------- +Time: 21.751s Load: 1.232s, Pack+Encode: 7.712s, Decode+Unpack: 12.807s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 330.3673 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01125-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01125-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst (17/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.251s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,745,148B, BPFP=1.1077 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,746,600B, BPFP=1.1087 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,107,028B, BPFP=1.3374 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,087,900B, BPFP=1.3253 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,247,620B, BPFP=1.4267 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,240,764B, BPFP=1.4223 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,471,836B, BPFP=0.9342 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,464,720B, BPFP=0.9297 +⌛️ [2/4] FRONTEND: Frontend time: 7.919s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.159s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00085355 4.41668243 + layer.9.1 0.00085581 4.41984380 + layer.19.0 0.00808159 15.44554990 + layer.19.1 0.00635426 11.13793366 + layer.29.0 4.24551200 36.68683224 + layer.29.1 4.24803037 30.98428360 + layer.39.0 9.19283951 1197.89478388 + layer.39.1 9.46657027 1226.04801755 + ------------------------------------------------------------------------------------- + TOTAL 3.39613717 315.87924088 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15111616 +BPFP 1.1990 bits/point +EBPFP 1.1990 equivalent bits/point +MSE 315.879241 +---------------------- --------------------------------------------------------- +Time: 22.329s Load: 1.251s, Pack+Encode: 7.919s, Decode+Unpack: 13.159s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 315.8792 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01135-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01135-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst (18/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.237s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,691,528B, BPFP=1.0737 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,709,588B, BPFP=1.0852 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,225,824B, BPFP=1.4128 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,227,808B, BPFP=1.4141 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,523,000B, BPFP=1.6015 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,525,756B, BPFP=1.6032 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,661,076B, BPFP=1.0544 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,649,524B, BPFP=1.0470 +⌛️ [2/4] FRONTEND: Frontend time: 7.847s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.103s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67056234 4.64324588 + layer.9.1 2.67147828 12.58286023 + layer.19.0 0.00618387 29.01848645 + layer.19.1 0.08383032 38.54730409 + layer.29.0 4.28489822 34.23628687 + layer.29.1 4.28470970 34.39748791 + layer.39.0 10.15376305 1119.24203770 + layer.39.1 8.47863686 1114.33839779 + ------------------------------------------------------------------------------------- + TOTAL 4.07925783 298.37576337 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16214104 +BPFP 1.2865 bits/point +EBPFP 1.2865 equivalent bits/point +MSE 298.375763 +---------------------- --------------------------------------------------------- +Time: 22.186s Load: 1.237s, Pack+Encode: 7.847s, Decode+Unpack: 13.103s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 298.3758 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01161-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01161-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst (19/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.267s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,678,784B, BPFP=1.0656 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,684,768B, BPFP=1.0694 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,194,108B, BPFP=1.3927 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,205,460B, BPFP=1.3999 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,468,200B, BPFP=1.5667 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,472,096B, BPFP=1.5692 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,606,016B, BPFP=1.0194 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,581,704B, BPFP=1.0040 +⌛️ [2/4] FRONTEND: Frontend time: 7.779s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.825s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67132404 4.38656482 + layer.9.1 2.67117709 4.37636471 + layer.19.0 0.00597838 25.04646622 + layer.19.1 0.00605309 38.22002966 + layer.29.0 4.29273040 53.87530976 + layer.29.1 4.29206328 87.63154046 + layer.39.0 9.96127074 1188.08896653 + layer.39.1 10.21295854 1276.58831654 + ------------------------------------------------------------------------------------- + TOTAL 4.26419445 334.77669484 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15891136 +BPFP 1.2609 bits/point +EBPFP 1.2609 equivalent bits/point +MSE 334.776695 +---------------------- --------------------------------------------------------- +Time: 21.871s Load: 1.267s, Pack+Encode: 7.779s, Decode+Unpack: 12.825s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 334.7767 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01162-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01162-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst (20/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.247s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,725,632B, BPFP=1.0953 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,732,448B, BPFP=1.0997 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,106,660B, BPFP=1.3372 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,113,192B, BPFP=1.3413 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,218,164B, BPFP=1.4080 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,224,712B, BPFP=1.4121 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,482,928B, BPFP=0.9413 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,485,208B, BPFP=0.9427 +⌛️ [2/4] FRONTEND: Frontend time: 7.925s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.090s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14568686 4.35059895 + layer.9.1 0.14558674 4.34653211 + layer.19.0 0.00960369 20.32962428 + layer.19.1 0.03847206 20.22137025 + layer.29.0 4.24438723 30.71106191 + layer.29.1 4.24578970 25.63481577 + layer.39.0 9.23757985 1315.29574261 + layer.39.1 9.43674592 1323.99618135 + ------------------------------------------------------------------------------------- + TOTAL 3.43798151 343.11074090 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15088944 +BPFP 1.1972 bits/point +EBPFP 1.1972 equivalent bits/point +MSE 343.110741 +---------------------- --------------------------------------------------------- +Time: 22.263s Load: 1.247s, Pack+Encode: 7.925s, Decode+Unpack: 13.090s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 343.1107 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01163-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01163-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst (21/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.241s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,789,588B, BPFP=1.1359 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,783,692B, BPFP=1.1322 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,317,396B, BPFP=1.4710 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,305,024B, BPFP=1.4631 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,588,672B, BPFP=1.6432 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,572,908B, BPFP=1.6332 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,654,712B, BPFP=1.0503 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,641,004B, BPFP=1.0416 +⌛️ [2/4] FRONTEND: Frontend time: 7.825s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.295s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11307967 4.38587453 + layer.9.1 0.00073224 4.49544408 + layer.19.0 0.08207503 20.55796154 + layer.19.1 0.08214869 29.71843263 + layer.29.0 4.26728487 63.07694183 + layer.29.1 4.26774951 18.51125030 + layer.39.0 12.81553410 1092.00463113 + layer.39.1 23.05196315 1150.80086123 + ------------------------------------------------------------------------------------- + TOTAL 5.58507091 297.94392466 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16652996 +BPFP 1.3213 bits/point +EBPFP 1.3213 equivalent bits/point +MSE 297.943925 +---------------------- --------------------------------------------------------- +Time: 22.361s Load: 1.241s, Pack+Encode: 7.825s, Decode+Unpack: 13.295s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 297.9439 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bedroom-rgb_01195-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bedroom-rgb_01195-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst (22/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.238s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,976,176B, BPFP=1.2544 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,977,884B, BPFP=1.2555 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,493,592B, BPFP=1.5828 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,495,656B, BPFP=1.5841 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,810,680B, BPFP=1.7841 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,813,084B, BPFP=1.7856 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,749,748B, BPFP=1.1107 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,740,688B, BPFP=1.1049 +⌛️ [2/4] FRONTEND: Frontend time: 7.308s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.671s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14519844 34.03012268 + layer.9.1 0.14499054 26.22621974 + layer.19.0 0.12156012 147.44297408 + layer.19.1 0.12030756 160.18633206 + layer.29.0 0.12020218 43.09596909 + layer.29.1 0.12115470 77.77423728 + layer.39.0 8.85439666 1174.32767306 + layer.39.1 8.75438231 1124.08425414 + ------------------------------------------------------------------------------------- + TOTAL 2.29777406 348.39597277 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 18057508 +BPFP 1.4327 bits/point +EBPFP 1.4327 equivalent bits/point +MSE 348.395973 +---------------------- --------------------------------------------------------- +Time: 21.217s Load: 1.238s, Pack+Encode: 7.308s, Decode+Unpack: 12.671s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 348.3960 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00083-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00083-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst (23/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.137s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,977,464B, BPFP=1.2552 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,975,556B, BPFP=1.2540 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,488,652B, BPFP=1.5797 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,478,124B, BPFP=1.5730 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,775,184B, BPFP=1.7615 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,760,196B, BPFP=1.7520 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,653,968B, BPFP=1.0499 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,649,980B, BPFP=1.0473 +⌛️ [2/4] FRONTEND: Frontend time: 7.762s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.151s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14541288 47.63241388 + layer.9.1 0.14479464 70.69911846 + layer.19.0 0.11855170 151.08921027 + layer.19.1 0.11778439 192.14295580 + layer.29.0 0.12648388 62.75503230 + layer.29.1 0.12520221 38.04998781 + layer.39.0 8.37129624 1058.66688333 + layer.39.1 8.45478741 1035.38674033 + ------------------------------------------------------------------------------------- + TOTAL 2.20053917 332.05279277 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17759124 +BPFP 1.4091 bits/point +EBPFP 1.4091 equivalent bits/point +MSE 332.052793 +---------------------- --------------------------------------------------------- +Time: 22.050s Load: 1.137s, Pack+Encode: 7.762s, Decode+Unpack: 13.151s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 332.0528 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00084-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00084-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst (24/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.163s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 2,035,748B, BPFP=1.2922 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 2,045,696B, BPFP=1.2985 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,592,232B, BPFP=1.6454 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,592,888B, BPFP=1.6458 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,926,444B, BPFP=1.8576 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,928,064B, BPFP=1.8586 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,945,556B, BPFP=1.2349 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,956,152B, BPFP=1.2417 +⌛️ [2/4] FRONTEND: Frontend time: 7.764s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.840s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14458197 16.80638101 + layer.9.1 0.14461228 45.31490697 + layer.19.0 0.12127609 168.38502397 + layer.19.1 0.12505172 150.03714048 + layer.29.0 0.11568762 42.93130484 + layer.29.1 0.11796058 58.23050557 + layer.39.0 8.63782956 1108.72676308 + layer.39.1 8.69862780 1166.17240819 + ------------------------------------------------------------------------------------- + TOTAL 2.26320345 344.57555426 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 19022780 +BPFP 1.5093 bits/point +EBPFP 1.5093 equivalent bits/point +MSE 344.575554 +---------------------- --------------------------------------------------------- +Time: 21.767s Load: 1.163s, Pack+Encode: 7.764s, Decode+Unpack: 12.840s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 344.5756 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00085-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00085-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst (25/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.269s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 2,038,920B, BPFP=1.2942 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 2,037,600B, BPFP=1.2934 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,602,296B, BPFP=1.6518 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,602,948B, BPFP=1.6522 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,933,044B, BPFP=1.8617 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,932,724B, BPFP=1.8615 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 2,000,712B, BPFP=1.2700 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 2,026,908B, BPFP=1.2866 +⌛️ [2/4] FRONTEND: Frontend time: 7.742s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.865s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14478087 33.84137350 + layer.9.1 0.14472154 20.93242708 + layer.19.0 0.13423899 164.89476357 + layer.19.1 0.13534726 160.16038349 + layer.29.0 0.11251127 52.43554497 + layer.29.1 0.11242151 42.67727291 + layer.39.0 10.58490794 1213.39250894 + layer.39.1 8.80008176 1193.82791680 + ------------------------------------------------------------------------------------- + TOTAL 2.52112639 360.27027391 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 19175152 +BPFP 1.5214 bits/point +EBPFP 1.5214 equivalent bits/point +MSE 360.270274 +---------------------- --------------------------------------------------------- +Time: 21.877s Load: 1.269s, Pack+Encode: 7.742s, Decode+Unpack: 12.865s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 360.2703 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00086-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00086-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst (26/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.265s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,925,248B, BPFP=1.2221 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,924,504B, BPFP=1.2216 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,424,852B, BPFP=1.5392 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,423,556B, BPFP=1.5384 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,737,000B, BPFP=1.7373 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,732,512B, BPFP=1.7345 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,818,108B, BPFP=1.1540 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,811,408B, BPFP=1.1498 +⌛️ [2/4] FRONTEND: Frontend time: 7.713s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.903s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14590075 34.31266250 + layer.9.1 0.14620647 58.39138670 + layer.19.0 0.11628058 38.74556437 + layer.19.1 0.11601873 57.13298769 + layer.29.0 0.11558260 61.24265721 + layer.29.1 0.11828149 61.26092785 + layer.39.0 28.43028163 1399.52876178 + layer.39.1 24.81181701 1462.81767956 + ------------------------------------------------------------------------------------- + TOTAL 6.75004616 396.67907846 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17797188 +BPFP 1.4121 bits/point +EBPFP 1.4121 equivalent bits/point +MSE 396.679078 +---------------------- --------------------------------------------------------- +Time: 21.882s Load: 1.265s, Pack+Encode: 7.713s, Decode+Unpack: 12.903s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 396.6791 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00090-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00090-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst (27/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.221s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,897,564B, BPFP=1.2045 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,907,760B, BPFP=1.2110 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,466,236B, BPFP=1.5654 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,458,780B, BPFP=1.5607 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,816,636B, BPFP=1.7879 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,803,740B, BPFP=1.7797 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,875,224B, BPFP=1.1903 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,850,280B, BPFP=1.1745 +⌛️ [2/4] FRONTEND: Frontend time: 7.596s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.878s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11529700 8.70728007 + layer.9.1 0.14629077 16.71630114 + layer.19.0 0.09721754 71.69035383 + layer.19.1 0.12446257 71.41194244 + layer.29.0 4.28687864 55.81406910 + layer.29.1 4.28715508 69.97458462 + layer.39.0 11.34089363 1252.04988625 + layer.39.1 19.75513766 1297.77429314 + ------------------------------------------------------------------------------------- + TOTAL 5.01916661 355.51733882 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 18076220 +BPFP 1.4342 bits/point +EBPFP 1.4342 equivalent bits/point +MSE 355.517339 +---------------------- --------------------------------------------------------- +Time: 21.694s Load: 1.221s, Pack+Encode: 7.596s, Decode+Unpack: 12.878s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 355.5173 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/bookstore-rgb_00116-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/bookstore-rgb_00116-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst (28/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.272s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,824,272B, BPFP=1.1580 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,835,084B, BPFP=1.1648 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,345,588B, BPFP=1.4889 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,351,056B, BPFP=1.4923 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,659,540B, BPFP=1.6881 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,668,836B, BPFP=1.6940 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,828,628B, BPFP=1.1607 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,856,228B, BPFP=1.1782 +⌛️ [2/4] FRONTEND: Frontend time: 7.883s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.789s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14538718 16.63893657 + layer.9.1 0.14538559 4.60599750 + layer.19.0 0.11434236 108.72556467 + layer.19.1 0.11406084 90.18549927 + layer.29.0 0.11219077 35.32780763 + layer.29.1 0.11281304 79.51823509 + layer.39.0 79.88316542 1456.90055249 + layer.39.1 46.71980622 1490.59132272 + ------------------------------------------------------------------------------------- + TOTAL 15.91839393 410.31173949 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17369232 +BPFP 1.3781 bits/point +EBPFP 1.3781 equivalent bits/point +MSE 410.311739 +---------------------- --------------------------------------------------------- +Time: 21.944s Load: 1.272s, Pack+Encode: 7.883s, Decode+Unpack: 12.789s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 410.3117 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00314-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00314-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst (29/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.264s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,900,736B, BPFP=1.2065 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,899,976B, BPFP=1.2060 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,399,444B, BPFP=1.5230 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,392,164B, BPFP=1.5184 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,714,440B, BPFP=1.7230 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,701,488B, BPFP=1.7148 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,884,500B, BPFP=1.1962 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,889,360B, BPFP=1.1993 +⌛️ [2/4] FRONTEND: Frontend time: 7.720s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.965s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14512126 21.57830527 + layer.9.1 0.14517278 25.95617942 + layer.19.0 0.11689420 165.27160180 + layer.19.1 0.12099910 110.48700032 + layer.29.0 0.11847120 80.72512797 + layer.29.1 0.12399357 75.63765437 + layer.39.0 75.86630139 1491.77868053 + layer.39.1 56.61936342 1490.10740981 + ------------------------------------------------------------------------------------- + TOTAL 16.65703962 432.69274494 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17782108 +BPFP 1.4109 bits/point +EBPFP 1.4109 equivalent bits/point +MSE 432.692745 +---------------------- --------------------------------------------------------- +Time: 21.948s Load: 1.264s, Pack+Encode: 7.720s, Decode+Unpack: 12.965s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 432.6927 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00316-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00316-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst (30/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.263s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,849,860B, BPFP=1.1742 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,838,656B, BPFP=1.1671 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,399,148B, BPFP=1.5229 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,388,432B, BPFP=1.5161 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,699,956B, BPFP=1.7138 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,690,104B, BPFP=1.7075 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,911,080B, BPFP=1.2131 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,903,608B, BPFP=1.2083 +⌛️ [2/4] FRONTEND: Frontend time: 7.679s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.662s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11488649 8.72246836 + layer.9.1 0.14606862 8.96508369 + layer.19.0 0.08767178 80.95992952 + layer.19.1 0.11443626 76.83306589 + layer.29.0 0.10933029 64.73860497 + layer.29.1 0.10817130 49.85151426 + layer.39.0 52.66717785 1465.89600260 + layer.39.1 62.91127214 1500.30110497 + ------------------------------------------------------------------------------------- + TOTAL 14.53237684 407.03347178 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17680844 +BPFP 1.4029 bits/point +EBPFP 1.4029 equivalent bits/point +MSE 407.033472 +---------------------- --------------------------------------------------------- +Time: 21.604s Load: 1.263s, Pack+Encode: 7.679s, Decode+Unpack: 12.662s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 407.0335 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00324-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00324-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst (31/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.264s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,896,060B, BPFP=1.2035 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,893,484B, BPFP=1.2019 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,371,256B, BPFP=1.5052 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,369,652B, BPFP=1.5041 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,654,952B, BPFP=1.6852 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,651,928B, BPFP=1.6833 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,853,480B, BPFP=1.1765 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,871,916B, BPFP=1.1882 +⌛️ [2/4] FRONTEND: Frontend time: 7.646s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.775s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14509552 17.43878600 + layer.9.1 0.14520687 9.18445574 + layer.19.0 0.12118574 76.73821397 + layer.19.1 0.11709642 77.00117302 + layer.29.0 0.10963326 85.44329907 + layer.29.1 0.10842036 55.21919686 + layer.39.0 53.79489966 1444.92655184 + layer.39.1 62.27410526 1375.88544036 + ------------------------------------------------------------------------------------- + TOTAL 14.60195539 392.72963961 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17562728 +BPFP 1.3935 bits/point +EBPFP 1.3935 equivalent bits/point +MSE 392.729640 +---------------------- --------------------------------------------------------- +Time: 21.685s Load: 1.264s, Pack+Encode: 7.646s, Decode+Unpack: 12.775s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 392.7296 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00325-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00325-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst (32/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.264s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 2,025,092B, BPFP=1.2854 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 2,020,308B, BPFP=1.2824 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,493,260B, BPFP=1.5826 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,491,440B, BPFP=1.5814 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,775,840B, BPFP=1.7620 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,773,908B, BPFP=1.7607 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,922,940B, BPFP=1.2206 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,918,552B, BPFP=1.2178 +⌛️ [2/4] FRONTEND: Frontend time: 7.709s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.918s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14577075 24.98092196 + layer.9.1 0.14541274 17.68463855 + layer.19.0 0.13069581 196.86135034 + layer.19.1 0.13545482 187.03593191 + layer.29.0 0.11331055 64.58526974 + layer.29.1 0.11244963 49.85557158 + layer.39.0 32.27446072 1301.16542086 + layer.39.1 16.59366367 1294.85066623 + ------------------------------------------------------------------------------------- + TOTAL 6.20640234 392.12747140 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 18421340 +BPFP 1.4616 bits/point +EBPFP 1.4616 equivalent bits/point +MSE 392.127471 +---------------------- --------------------------------------------------------- +Time: 21.891s Load: 1.264s, Pack+Encode: 7.709s, Decode+Unpack: 12.918s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 392.1275 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00327-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00327-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst (33/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.222s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,823,452B, BPFP=1.1574 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,852,052B, BPFP=1.1756 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,298,216B, BPFP=1.4588 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,294,152B, BPFP=1.4562 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,598,868B, BPFP=1.6496 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,599,388B, BPFP=1.6500 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,785,244B, BPFP=1.1332 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,795,568B, BPFP=1.1397 +⌛️ [2/4] FRONTEND: Frontend time: 7.642s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.876s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14585920 8.68547388 + layer.9.1 0.14576220 24.74923830 + layer.19.0 0.12270736 76.30080029 + layer.19.1 0.12453605 113.40503128 + layer.29.0 0.11393550 99.56625772 + layer.29.1 0.11678154 108.42169727 + layer.39.0 53.83016636 1315.40396490 + layer.39.1 40.65720720 1286.87211570 + ------------------------------------------------------------------------------------- + TOTAL 11.90711942 379.17557242 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17046940 +BPFP 1.3526 bits/point +EBPFP 1.3526 equivalent bits/point +MSE 379.175572 +---------------------- --------------------------------------------------------- +Time: 21.740s Load: 1.222s, Pack+Encode: 7.642s, Decode+Unpack: 12.876s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 379.1756 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/classroom-rgb_00328-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/classroom-rgb_00328-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst (34/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.115s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,769,348B, BPFP=1.1231 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,773,976B, BPFP=1.1260 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,348,068B, BPFP=1.4904 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,349,636B, BPFP=1.4914 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,672,184B, BPFP=1.6962 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,676,028B, BPFP=1.6986 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,725,728B, BPFP=1.0954 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,748,948B, BPFP=1.1101 +⌛️ [2/4] FRONTEND: Frontend time: 7.585s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.238s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14568599 4.34195810 + layer.9.1 0.03329684 20.71328607 + layer.19.0 0.11848472 53.33659002 + layer.19.1 0.11973745 48.35107755 + layer.29.0 0.10886538 77.98953932 + layer.29.1 0.10946879 43.83863138 + layer.39.0 14.08931437 1174.08969776 + layer.39.1 9.95616799 1197.35919727 + ------------------------------------------------------------------------------------- + TOTAL 3.08512769 327.50249718 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17063916 +BPFP 1.3539 bits/point +EBPFP 1.3539 equivalent bits/point +MSE 327.502497 +---------------------- --------------------------------------------------------- +Time: 21.938s Load: 1.115s, Pack+Encode: 7.585s, Decode+Unpack: 13.238s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 327.5025 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00332-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00332-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst (35/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.154s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,863,028B, BPFP=1.1826 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,852,164B, BPFP=1.1757 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,331,324B, BPFP=1.4798 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,338,020B, BPFP=1.4841 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,623,008B, BPFP=1.6650 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,624,252B, BPFP=1.6657 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,745,780B, BPFP=1.1081 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,757,360B, BPFP=1.1155 +⌛️ [2/4] FRONTEND: Frontend time: 7.761s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.842s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14493624 16.98267006 + layer.9.1 0.14482686 16.56486635 + layer.19.0 0.11946148 84.95450114 + layer.19.1 0.12828579 76.40988077 + layer.29.0 0.10467725 88.98490819 + layer.29.1 0.10613328 39.95442497 + layer.39.0 22.00188902 1256.61382840 + layer.39.1 19.26198661 1255.84286643 + ------------------------------------------------------------------------------------- + TOTAL 5.25152457 354.53849329 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17134936 +BPFP 1.3595 bits/point +EBPFP 1.3595 equivalent bits/point +MSE 354.538493 +---------------------- --------------------------------------------------------- +Time: 21.756s Load: 1.154s, Pack+Encode: 7.761s, Decode+Unpack: 12.842s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 354.5385 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00333-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00333-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst (36/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.274s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,873,820B, BPFP=1.1894 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,874,456B, BPFP=1.1898 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,325,700B, BPFP=1.4762 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,322,220B, BPFP=1.4740 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,587,448B, BPFP=1.6424 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,569,048B, BPFP=1.6307 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,739,416B, BPFP=1.1041 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,718,288B, BPFP=1.0907 +⌛️ [2/4] FRONTEND: Frontend time: 7.877s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.826s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14475438 12.98277416 + layer.9.1 0.14492096 8.64440493 + layer.19.0 0.11744098 47.67351519 + layer.19.1 0.11578254 52.96885664 + layer.29.0 0.11402616 99.05692436 + layer.29.1 0.11062706 133.41255687 + layer.39.0 28.92800668 1369.36366591 + layer.39.1 10.80449708 1356.52575561 + ------------------------------------------------------------------------------------- + TOTAL 5.06000698 385.07855671 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17010396 +BPFP 1.3497 bits/point +EBPFP 1.3497 equivalent bits/point +MSE 385.078557 +---------------------- --------------------------------------------------------- +Time: 21.977s Load: 1.274s, Pack+Encode: 7.877s, Decode+Unpack: 12.826s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 385.0786 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/computer_lab-rgb_00334-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/computer_lab-rgb_00334-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst (37/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.265s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,753,200B, BPFP=1.1128 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,743,112B, BPFP=1.1064 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,150,672B, BPFP=1.3651 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,205,944B, BPFP=1.4002 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,346,336B, BPFP=1.4893 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,356,204B, BPFP=1.4956 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,549,796B, BPFP=0.9837 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,544,816B, BPFP=0.9806 +⌛️ [2/4] FRONTEND: Frontend time: 7.587s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.153s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14546214 8.40357834 + layer.9.1 0.14553630 4.40373163 + layer.19.0 0.04765745 25.40438384 + layer.19.1 0.04191649 25.58009526 + layer.29.0 0.16505912 212.25249838 + layer.29.1 0.15755973 235.43628128 + layer.39.0 42.51041751 1298.27941176 + layer.39.1 31.38856333 1309.90201495 + ------------------------------------------------------------------------------------- + TOTAL 9.32527151 389.95774943 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15650080 +BPFP 1.2417 bits/point +EBPFP 1.2417 equivalent bits/point +MSE 389.957749 +---------------------- --------------------------------------------------------- +Time: 22.005s Load: 1.265s, Pack+Encode: 7.587s, Decode+Unpack: 13.153s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 389.9577 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01347-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01347-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst (38/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.193s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,733,752B, BPFP=1.1005 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,753,560B, BPFP=1.1131 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,287,916B, BPFP=1.4523 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,304,512B, BPFP=1.4628 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,528,956B, BPFP=1.6053 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,532,416B, BPFP=1.6075 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,686,060B, BPFP=1.0702 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,698,752B, BPFP=1.0783 +⌛️ [2/4] FRONTEND: Frontend time: 7.259s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.941s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03343166 12.41381749 + layer.9.1 0.03311388 16.52544966 + layer.19.0 0.03842411 19.79101626 + layer.19.1 0.03806642 20.26102814 + layer.29.0 4.26870163 52.61059880 + layer.29.1 4.26552788 38.65788765 + layer.39.0 33.95300821 1051.67549561 + layer.39.1 48.19954501 1023.32539812 + ------------------------------------------------------------------------------------- + TOTAL 11.35372735 279.40758647 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16525924 +BPFP 1.3112 bits/point +EBPFP 1.3112 equivalent bits/point +MSE 279.407586 +---------------------- --------------------------------------------------------- +Time: 21.393s Load: 1.193s, Pack+Encode: 7.259s, Decode+Unpack: 12.941s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 279.4076 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01390-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01390-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst (39/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.178s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,730,972B, BPFP=1.0987 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,735,608B, BPFP=1.1017 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,274,572B, BPFP=1.4438 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,257,552B, BPFP=1.4330 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,482,192B, BPFP=1.5756 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,481,200B, BPFP=1.5749 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,650,488B, BPFP=1.0476 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,688,068B, BPFP=1.0715 +⌛️ [2/4] FRONTEND: Frontend time: 7.170s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.746s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14549638 16.85127559 + layer.9.1 0.14520178 12.80669585 + layer.19.0 0.11487435 47.85823245 + layer.19.1 0.11481158 34.06503646 + layer.29.0 0.10827909 39.00327023 + layer.29.1 0.10618535 68.77215023 + layer.39.0 9.83978281 1145.34579136 + layer.39.1 9.67554703 1145.47123822 + ------------------------------------------------------------------------------------- + TOTAL 2.53127230 313.77171130 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16300652 +BPFP 1.2934 bits/point +EBPFP 1.2934 equivalent bits/point +MSE 313.771711 +---------------------- --------------------------------------------------------- +Time: 21.093s Load: 1.178s, Pack+Encode: 7.170s, Decode+Unpack: 12.746s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 313.7717 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01413-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01413-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst (40/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.089s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,784,192B, BPFP=1.1325 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,789,592B, BPFP=1.1359 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,268,380B, BPFP=1.4399 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,259,388B, BPFP=1.4341 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,556,476B, BPFP=1.6227 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,553,336B, BPFP=1.6207 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,630,548B, BPFP=1.0350 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,634,368B, BPFP=1.0374 +⌛️ [2/4] FRONTEND: Frontend time: 7.837s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.389s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00092318 4.37109438 + layer.9.1 0.00095285 8.58443632 + layer.19.0 0.08568402 24.50849549 + layer.19.1 0.08404610 48.56978185 + layer.29.0 0.12100375 69.58037455 + layer.29.1 0.12795564 59.83608222 + layer.39.0 12.85620633 1280.52193695 + layer.39.1 12.98640239 1329.01998700 + ------------------------------------------------------------------------------------- + TOTAL 3.28289678 353.12402360 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16476280 +BPFP 1.3073 bits/point +EBPFP 1.3073 equivalent bits/point +MSE 353.124024 +---------------------- --------------------------------------------------------- +Time: 22.315s Load: 1.089s, Pack+Encode: 7.837s, Decode+Unpack: 13.389s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 353.1240 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01421-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01421-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst (41/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.119s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,797,564B, BPFP=1.1410 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,802,148B, BPFP=1.1439 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,304,972B, BPFP=1.4631 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,299,152B, BPFP=1.4594 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,607,876B, BPFP=1.6553 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,590,384B, BPFP=1.6442 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,715,984B, BPFP=1.0892 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,723,388B, BPFP=1.0939 +⌛️ [2/4] FRONTEND: Frontend time: 7.559s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.838s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00113376 8.72097480 + layer.9.1 0.00100095 4.50901789 + layer.19.0 0.00983371 42.70323164 + layer.19.1 0.00806405 24.61187845 + layer.29.0 4.28365570 29.46446671 + layer.29.1 4.28597952 38.74091800 + layer.39.0 8.41906814 1156.79468638 + layer.39.1 8.59662605 1130.70726357 + ------------------------------------------------------------------------------------- + TOTAL 3.20067024 304.53155468 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16841468 +BPFP 1.3363 bits/point +EBPFP 1.3363 equivalent bits/point +MSE 304.531555 +---------------------- --------------------------------------------------------- +Time: 21.516s Load: 1.119s, Pack+Encode: 7.559s, Decode+Unpack: 12.838s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 304.5316 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01422-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01422-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst (42/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.208s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,883,904B, BPFP=1.1958 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,861,604B, BPFP=1.1817 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,356,064B, BPFP=1.4955 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,356,644B, BPFP=1.4959 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,651,404B, BPFP=1.6830 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,654,964B, BPFP=1.6852 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,726,156B, BPFP=1.0957 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,737,236B, BPFP=1.1027 +⌛️ [2/4] FRONTEND: Frontend time: 7.901s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.725s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14550294 16.88008054 + layer.9.1 0.14526658 5.13031825 + layer.19.0 0.11599200 15.12517646 + layer.19.1 0.11361485 20.11097330 + layer.29.0 4.26439454 34.33846634 + layer.29.1 4.25587461 39.29639513 + layer.39.0 8.37236706 1092.93443289 + layer.39.1 8.35116642 1093.23675658 + ------------------------------------------------------------------------------------- + TOTAL 3.22052237 289.63157494 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17227976 +BPFP 1.3669 bits/point +EBPFP 1.3669 equivalent bits/point +MSE 289.631575 +---------------------- --------------------------------------------------------- +Time: 21.834s Load: 1.208s, Pack+Encode: 7.901s, Decode+Unpack: 12.725s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 289.6316 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/dining_room-rgb_01441-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/dining_room-rgb_01441-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst (43/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.274s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,811,916B, BPFP=1.1501 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,835,612B, BPFP=1.1652 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,366,060B, BPFP=1.5019 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,372,248B, BPFP=1.5058 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,655,964B, BPFP=1.6859 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,661,392B, BPFP=1.6893 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,743,936B, BPFP=1.1070 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,756,424B, BPFP=1.1149 +⌛️ [2/4] FRONTEND: Frontend time: 7.733s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.725s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00080404 4.51916627 + layer.9.1 0.00082438 4.37608320 + layer.19.0 0.00843097 24.79355653 + layer.19.1 0.00674472 52.18203607 + layer.29.0 4.27713270 48.88280488 + layer.29.1 4.27133426 29.52296017 + layer.39.0 22.97048921 1116.36675333 + layer.39.1 18.06488920 1089.01722457 + ------------------------------------------------------------------------------------- + TOTAL 6.20008118 296.20757313 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17203552 +BPFP 1.3650 bits/point +EBPFP 1.3650 equivalent bits/point +MSE 296.207573 +---------------------- --------------------------------------------------------- +Time: 21.732s Load: 1.274s, Pack+Encode: 7.733s, Decode+Unpack: 12.725s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 296.2076 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00350-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00350-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst (44/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.201s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,818,616B, BPFP=1.1544 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,830,168B, BPFP=1.1617 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,344,980B, BPFP=1.4885 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,354,668B, BPFP=1.4946 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,640,416B, BPFP=1.6760 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,648,376B, BPFP=1.6811 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,745,700B, BPFP=1.1081 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,749,256B, BPFP=1.1103 +⌛️ [2/4] FRONTEND: Frontend time: 7.276s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.972s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14506060 24.61219075 + layer.9.1 0.14523201 8.61444729 + layer.19.0 0.04621643 33.83412719 + layer.19.1 0.04629335 33.90474437 + layer.29.0 4.27940669 53.85539893 + layer.29.1 4.27759670 83.91593374 + layer.39.0 19.91382637 1097.96124472 + layer.39.1 24.01088215 1116.40697108 + ------------------------------------------------------------------------------------- + TOTAL 6.60806429 306.63813226 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17132180 +BPFP 1.3593 bits/point +EBPFP 1.3593 equivalent bits/point +MSE 306.638132 +---------------------- --------------------------------------------------------- +Time: 21.449s Load: 1.201s, Pack+Encode: 7.276s, Decode+Unpack: 12.972s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 306.6381 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/foyer-rgb_00351-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/foyer-rgb_00351-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst (45/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.116s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,666,528B, BPFP=1.0578 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,669,472B, BPFP=1.0597 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,043,144B, BPFP=1.2969 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,030,760B, BPFP=1.2890 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,189,868B, BPFP=1.3900 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,198,644B, BPFP=1.3956 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,430,896B, BPFP=0.9083 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,436,308B, BPFP=0.9117 +⌛️ [2/4] FRONTEND: Frontend time: 7.749s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.893s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66985745 4.40762169 + layer.9.1 2.66884121 4.53968582 + layer.19.0 3.21935619 15.24819604 + layer.19.1 3.21606501 6.01491472 + layer.29.0 4.24164606 35.40495765 + layer.29.1 4.23648681 54.67647567 + layer.39.0 8.06392628 1183.70230744 + layer.39.1 8.17747540 1217.07710432 + ------------------------------------------------------------------------------------- + TOTAL 4.56170680 315.13390792 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 14665620 +BPFP 1.1636 bits/point +EBPFP 1.1636 equivalent bits/point +MSE 315.133908 +---------------------- --------------------------------------------------------- +Time: 21.757s Load: 1.116s, Pack+Encode: 7.749s, Decode+Unpack: 12.893s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 315.1339 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00358-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00358-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst (46/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.267s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,697,008B, BPFP=1.0772 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,698,804B, BPFP=1.0783 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,078,284B, BPFP=1.3192 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,067,200B, BPFP=1.3122 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,321,872B, BPFP=1.4738 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,310,188B, BPFP=1.4664 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,584,872B, BPFP=1.0060 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,547,020B, BPFP=0.9820 +⌛️ [2/4] FRONTEND: Frontend time: 7.852s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.094s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66795352 4.68120487 + layer.9.1 2.66862889 4.62444840 + layer.19.0 3.22250645 29.15878392 + layer.19.1 3.22577319 15.72665263 + layer.29.0 4.25792136 49.56073793 + layer.29.1 4.25014663 36.02047449 + layer.39.0 8.65209937 1201.55037374 + layer.39.1 8.58450170 1219.07702307 + ------------------------------------------------------------------------------------- + TOTAL 4.69119139 320.04996238 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15305248 +BPFP 1.2144 bits/point +EBPFP 1.2144 equivalent bits/point +MSE 320.049962 +---------------------- --------------------------------------------------------- +Time: 22.213s Load: 1.267s, Pack+Encode: 7.852s, Decode+Unpack: 13.094s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 320.0500 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00360-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00360-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst (47/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.239s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,825,368B, BPFP=1.1587 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,801,356B, BPFP=1.1434 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,364,740B, BPFP=1.5010 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,366,772B, BPFP=1.5023 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,679,444B, BPFP=1.7008 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,680,024B, BPFP=1.7011 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,703,708B, BPFP=1.0814 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,685,164B, BPFP=1.0697 +⌛️ [2/4] FRONTEND: Frontend time: 7.856s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.721s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00092775 4.40924602 + layer.9.1 0.00093166 4.41329636 + layer.19.0 0.08227225 24.95655519 + layer.19.1 0.08381199 20.47876635 + layer.29.0 0.10725604 29.69845324 + layer.29.1 0.10756977 24.15545022 + layer.39.0 7.96294394 947.37495938 + layer.39.1 7.95922050 929.72359441 + ------------------------------------------------------------------------------------- + TOTAL 2.03811674 248.15129014 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17106576 +BPFP 1.3573 bits/point +EBPFP 1.3573 equivalent bits/point +MSE 248.151290 +---------------------- --------------------------------------------------------- +Time: 21.816s Load: 1.239s, Pack+Encode: 7.856s, Decode+Unpack: 12.721s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 248.1513 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00361-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00361-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst (48/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.267s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,703,904B, BPFP=1.0816 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,703,008B, BPFP=1.0810 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,153,976B, BPFP=1.3672 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,151,656B, BPFP=1.3658 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,351,780B, BPFP=1.4928 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,319,708B, BPFP=1.4724 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,510,128B, BPFP=0.9586 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,517,576B, BPFP=0.9633 +⌛️ [2/4] FRONTEND: Frontend time: 7.813s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.178s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66308118 4.45811509 + layer.9.1 2.66351027 4.44495704 + layer.19.0 3.21594155 5.94440227 + layer.19.1 3.21498593 10.66588677 + layer.29.0 4.33566519 163.49592745 + layer.29.1 4.34101296 168.22668183 + layer.39.0 8.65310735 1098.87658434 + layer.39.1 8.66575030 1120.97481313 + ------------------------------------------------------------------------------------- + TOTAL 4.71913184 322.13592099 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15411736 +BPFP 1.2228 bits/point +EBPFP 1.2228 equivalent bits/point +MSE 322.135921 +---------------------- --------------------------------------------------------- +Time: 22.259s Load: 1.267s, Pack+Encode: 7.813s, Decode+Unpack: 13.178s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 322.1359 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00384-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00384-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst (49/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.245s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,669,372B, BPFP=1.0596 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,675,652B, BPFP=1.0636 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,175,052B, BPFP=1.3806 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,179,564B, BPFP=1.3835 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,343,636B, BPFP=1.4876 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,344,472B, BPFP=1.4882 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,483,804B, BPFP=0.9418 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,447,796B, BPFP=0.9190 +⌛️ [2/4] FRONTEND: Frontend time: 7.684s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.614s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.65959259 4.45417646 + layer.9.1 2.65993726 4.57922788 + layer.19.0 3.20866700 10.53629372 + layer.19.1 3.21007805 19.86830085 + layer.29.0 4.27255361 121.03712017 + layer.29.1 4.27602442 169.30513690 + layer.39.0 19.11658068 1171.41257719 + layer.39.1 9.60360322 1147.52063698 + ------------------------------------------------------------------------------------- + TOTAL 6.12587960 331.08918377 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15319348 +BPFP 1.2155 bits/point +EBPFP 1.2155 equivalent bits/point +MSE 331.089184 +---------------------- --------------------------------------------------------- +Time: 21.543s Load: 1.245s, Pack+Encode: 7.684s, Decode+Unpack: 12.614s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 331.0892 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00385-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00385-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst (50/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.263s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,722,092B, BPFP=1.0931 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,710,220B, BPFP=1.0856 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,219,008B, BPFP=1.4085 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,214,820B, BPFP=1.4059 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,416,800B, BPFP=1.5341 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,404,108B, BPFP=1.5260 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,555,368B, BPFP=0.9873 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,558,136B, BPFP=0.9890 +⌛️ [2/4] FRONTEND: Frontend time: 7.813s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.005s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67078767 4.38779338 + layer.9.1 2.67131261 4.37765135 + layer.19.0 3.30595795 33.88506967 + layer.19.1 3.30543206 25.17476946 + layer.29.0 0.11228124 134.86055817 + layer.29.1 0.11507649 115.29859644 + layer.39.0 11.41791162 1176.15705232 + layer.39.1 11.38150745 1151.62138447 + ------------------------------------------------------------------------------------- + TOTAL 4.37253339 330.72035941 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15800552 +BPFP 1.2537 bits/point +EBPFP 1.2537 equivalent bits/point +MSE 330.720359 +---------------------- --------------------------------------------------------- +Time: 22.080s Load: 1.263s, Pack+Encode: 7.813s, Decode+Unpack: 13.005s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 330.7204 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/home_office-rgb_00557-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/home_office-rgb_00557-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst (51/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.246s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,781,276B, BPFP=1.1307 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,789,572B, BPFP=1.1359 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,224,828B, BPFP=1.4122 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,223,996B, BPFP=1.4117 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,501,496B, BPFP=1.5878 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,496,196B, BPFP=1.5845 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,637,880B, BPFP=1.0396 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,659,668B, BPFP=1.0535 +⌛️ [2/4] FRONTEND: Frontend time: 7.803s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.316s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14463971 4.41666434 + layer.9.1 0.14470460 8.59775400 + layer.19.0 0.12255537 38.45240139 + layer.19.1 0.11825690 38.57739377 + layer.29.0 0.11949990 106.58157296 + layer.29.1 0.11467140 100.93206654 + layer.39.0 10.68243977 1173.70417615 + layer.39.1 10.40156301 1094.77315567 + ------------------------------------------------------------------------------------- + TOTAL 2.73104133 320.75439810 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16314912 +BPFP 1.2945 bits/point +EBPFP 1.2945 equivalent bits/point +MSE 320.754398 +---------------------- --------------------------------------------------------- +Time: 22.365s Load: 1.246s, Pack+Encode: 7.803s, Decode+Unpack: 13.316s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 320.7544 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00001-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00001-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst (52/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.241s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,834,768B, BPFP=1.1646 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,832,588B, BPFP=1.1632 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,329,584B, BPFP=1.4787 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,311,232B, BPFP=1.4671 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,550,644B, BPFP=1.6190 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,544,920B, BPFP=1.6154 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,677,632B, BPFP=1.0649 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,663,932B, BPFP=1.0562 +⌛️ [2/4] FRONTEND: Frontend time: 7.603s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.692s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14485273 8.59844207 + layer.9.1 0.14484227 4.83212011 + layer.19.0 0.11969613 80.97800719 + layer.19.1 0.11916645 56.68613910 + layer.29.0 0.11480527 68.78806467 + layer.29.1 0.11451660 73.12023176 + layer.39.0 11.00270276 1187.81686708 + layer.39.1 11.01557422 1153.54387390 + ------------------------------------------------------------------------------------- + TOTAL 2.84701956 329.29546824 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16745300 +BPFP 1.3286 bits/point +EBPFP 1.3286 equivalent bits/point +MSE 329.295468 +---------------------- --------------------------------------------------------- +Time: 21.536s Load: 1.241s, Pack+Encode: 7.603s, Decode+Unpack: 12.692s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 329.2955 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00126-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00126-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst (53/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.239s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,803,296B, BPFP=1.1446 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,815,716B, BPFP=1.1525 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,171,292B, BPFP=1.3782 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,170,100B, BPFP=1.3775 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,311,472B, BPFP=1.4672 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,308,404B, BPFP=1.4653 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,607,456B, BPFP=1.0203 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,615,684B, BPFP=1.0256 +⌛️ [2/4] FRONTEND: Frontend time: 7.371s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.702s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14481488 8.82645624 + layer.9.1 0.14470567 4.42786609 + layer.19.0 0.03819180 48.93110172 + layer.19.1 0.04002141 67.25977515 + layer.29.0 0.11241068 105.50236635 + layer.29.1 0.11133552 62.32060956 + layer.39.0 31.78807483 1490.27201820 + layer.39.1 43.50691623 1444.64494638 + ------------------------------------------------------------------------------------- + TOTAL 9.48580888 404.02314246 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15803420 +BPFP 1.2539 bits/point +EBPFP 1.2539 equivalent bits/point +MSE 404.023142 +---------------------- --------------------------------------------------------- +Time: 21.313s Load: 1.239s, Pack+Encode: 7.371s, Decode+Unpack: 12.702s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 404.0231 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00127-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00127-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst (54/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.241s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,775,380B, BPFP=1.1269 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,759,544B, BPFP=1.1169 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,233,524B, BPFP=1.4177 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,242,868B, BPFP=1.4237 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,480,972B, BPFP=1.5748 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,499,220B, BPFP=1.5864 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,629,388B, BPFP=1.0343 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,649,204B, BPFP=1.0468 +⌛️ [2/4] FRONTEND: Frontend time: 7.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.669s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14528272 12.64162283 + layer.9.1 0.14516892 8.42806444 + layer.19.0 0.11319376 43.11073590 + layer.19.1 0.11666145 43.88318573 + layer.29.0 0.21118872 213.38147953 + layer.29.1 0.20646930 220.15130403 + layer.39.0 14.37750853 1494.44978876 + layer.39.1 21.76644002 1553.10464738 + ------------------------------------------------------------------------------------- + TOTAL 4.63523918 448.64385358 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16270100 +BPFP 1.2909 bits/point +EBPFP 1.2909 equivalent bits/point +MSE 448.643854 +---------------------- --------------------------------------------------------- +Time: 21.330s Load: 1.241s, Pack+Encode: 7.420s, Decode+Unpack: 12.669s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 448.6439 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00128-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00128-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst (55/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.244s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,737,152B, BPFP=1.1027 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,742,248B, BPFP=1.1059 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,219,188B, BPFP=1.4086 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,201,600B, BPFP=1.3975 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,426,060B, BPFP=1.5399 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,409,480B, BPFP=1.5294 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,654,564B, BPFP=1.0502 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,641,948B, BPFP=1.0422 +⌛️ [2/4] FRONTEND: Frontend time: 7.374s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.654s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14454765 12.73541980 + layer.9.1 0.14475082 12.53812688 + layer.19.0 0.04087094 25.65763375 + layer.19.1 0.11687931 38.53468780 + layer.29.0 0.10817139 54.06053481 + layer.29.1 0.10802081 78.32344309 + layer.39.0 19.80422286 1287.29338642 + layer.39.1 34.29222355 1249.04493013 + ------------------------------------------------------------------------------------- + TOTAL 6.84496092 344.77352033 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16032240 +BPFP 1.2721 bits/point +EBPFP 1.2721 equivalent bits/point +MSE 344.773520 +---------------------- --------------------------------------------------------- +Time: 21.272s Load: 1.244s, Pack+Encode: 7.374s, Decode+Unpack: 12.654s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 344.7735 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00131-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00131-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst (56/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.240s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,717,708B, BPFP=1.0903 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,713,236B, BPFP=1.0875 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,137,292B, BPFP=1.3566 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,156,808B, BPFP=1.3690 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,317,576B, BPFP=1.4711 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,338,140B, BPFP=1.4841 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,573,484B, BPFP=0.9988 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,587,620B, BPFP=1.0077 +⌛️ [2/4] FRONTEND: Frontend time: 7.370s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.645s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14462205 4.70416250 + layer.9.1 0.14495783 4.30640196 + layer.19.0 0.04322015 38.84281057 + layer.19.1 0.03788725 48.27695909 + layer.29.0 0.10021623 50.12525390 + layer.29.1 0.10137775 49.95849752 + layer.39.0 58.66958482 1271.68800780 + layer.39.1 72.48303949 1249.12300942 + ------------------------------------------------------------------------------------- + TOTAL 16.46561320 339.62813785 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15541864 +BPFP 1.2331 bits/point +EBPFP 1.2331 equivalent bits/point +MSE 339.628138 +---------------------- --------------------------------------------------------- +Time: 21.255s Load: 1.240s, Pack+Encode: 7.370s, Decode+Unpack: 12.645s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 339.6281 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00132-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00132-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst (57/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.245s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,842,144B, BPFP=1.1693 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,852,672B, BPFP=1.1760 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,349,692B, BPFP=1.4915 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,353,796B, BPFP=1.4941 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,591,312B, BPFP=1.6448 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,602,916B, BPFP=1.6522 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,639,280B, BPFP=1.0405 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,646,748B, BPFP=1.0453 +⌛️ [2/4] FRONTEND: Frontend time: 7.882s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.743s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14527927 17.02842282 + layer.9.1 0.14528875 4.57092123 + layer.19.0 0.12591341 30.09905651 + layer.19.1 0.13556211 30.11327998 + layer.29.0 0.11238900 76.93036541 + layer.29.1 0.11028371 124.82966363 + layer.39.0 11.48751193 1143.94946376 + layer.39.1 11.29491489 1153.54923627 + ------------------------------------------------------------------------------------- + TOTAL 2.94464289 322.63380120 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16878560 +BPFP 1.3392 bits/point +EBPFP 1.3392 equivalent bits/point +MSE 322.633801 +---------------------- --------------------------------------------------------- +Time: 21.869s Load: 1.245s, Pack+Encode: 7.882s, Decode+Unpack: 12.743s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 322.6338 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00136-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00136-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst (58/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.244s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,835,184B, BPFP=1.1649 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,837,204B, BPFP=1.1662 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,303,300B, BPFP=1.4620 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,296,068B, BPFP=1.4574 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,537,304B, BPFP=1.6106 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,537,828B, BPFP=1.6109 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,651,504B, BPFP=1.0483 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,609,468B, BPFP=1.0216 +⌛️ [2/4] FRONTEND: Frontend time: 7.238s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.621s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14489184 8.84808661 + layer.9.1 0.14511764 8.91491243 + layer.19.0 0.03976490 29.60113341 + layer.19.1 0.11370806 52.13622745 + layer.29.0 0.10933599 66.89265112 + layer.29.1 0.11012027 48.26307584 + layer.39.0 9.10787636 966.62390315 + layer.39.1 9.00026152 1013.29793630 + ------------------------------------------------------------------------------------- + TOTAL 2.34638457 274.32224079 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16607860 +BPFP 1.3177 bits/point +EBPFP 1.3177 equivalent bits/point +MSE 274.322241 +---------------------- --------------------------------------------------------- +Time: 21.103s Load: 1.244s, Pack+Encode: 7.238s, Decode+Unpack: 12.621s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 274.3222 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00193-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00193-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst (59/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.278s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,742,216B, BPFP=1.1059 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,746,096B, BPFP=1.1083 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,209,008B, BPFP=1.4022 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,201,084B, BPFP=1.3971 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,334,004B, BPFP=1.4815 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,331,012B, BPFP=1.4796 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,408,856B, BPFP=0.8943 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,387,548B, BPFP=0.8807 +⌛️ [2/4] FRONTEND: Frontend time: 7.366s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.640s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00083307 4.35174880 + layer.9.1 0.00247171 4.34822721 + layer.19.0 0.00642632 10.50450165 + layer.19.1 0.00641681 24.76115128 + layer.29.0 0.10256791 29.09348848 + layer.29.1 0.10162673 23.74429995 + layer.39.0 8.50517638 1013.10261618 + layer.39.1 8.55767781 972.36236594 + ------------------------------------------------------------------------------------- + TOTAL 2.16039959 260.28354994 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15359824 +BPFP 1.2187 bits/point +EBPFP 1.2187 equivalent bits/point +MSE 260.283550 +---------------------- --------------------------------------------------------- +Time: 21.284s Load: 1.278s, Pack+Encode: 7.366s, Decode+Unpack: 12.640s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 260.2835 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00196-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00196-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst (60/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.243s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,749,972B, BPFP=1.1108 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,757,880B, BPFP=1.1158 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,209,352B, BPFP=1.4024 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,212,564B, BPFP=1.4044 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,514,116B, BPFP=1.5958 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,509,344B, BPFP=1.5928 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,638,780B, BPFP=1.0402 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,620,896B, BPFP=1.0289 +⌛️ [2/4] FRONTEND: Frontend time: 7.385s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.676s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00064391 4.58792744 + layer.9.1 0.00065402 4.36700818 + layer.19.0 0.08134466 15.70038796 + layer.19.1 0.08141702 24.34308986 + layer.29.0 0.11551180 101.46231110 + layer.29.1 0.11251285 91.33017143 + layer.39.0 10.61319619 1137.17297693 + layer.39.1 10.43102047 1159.72513812 + ------------------------------------------------------------------------------------- + TOTAL 2.67953762 317.33612638 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16212904 +BPFP 1.2864 bits/point +EBPFP 1.2864 equivalent bits/point +MSE 317.336126 +---------------------- --------------------------------------------------------- +Time: 21.303s Load: 1.243s, Pack+Encode: 7.385s, Decode+Unpack: 12.676s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 317.3361 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00559-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00559-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst (61/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.244s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,795,324B, BPFP=1.1396 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,777,168B, BPFP=1.1281 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,270,788B, BPFP=1.4414 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,269,620B, BPFP=1.4406 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,567,696B, BPFP=1.6298 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,576,032B, BPFP=1.6351 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,725,880B, BPFP=1.0955 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,752,016B, BPFP=1.1121 +⌛️ [2/4] FRONTEND: Frontend time: 7.280s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.659s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14441776 8.54027932 + layer.9.1 0.14449203 17.03794026 + layer.19.0 0.11315974 47.89674907 + layer.19.1 0.11435745 57.36772323 + layer.29.0 0.12811458 150.40930695 + layer.29.1 0.12952277 115.93622034 + layer.39.0 31.10682331 1297.21530712 + layer.39.1 16.99297713 1259.88438414 + ------------------------------------------------------------------------------------- + TOTAL 6.10923309 369.28598880 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16734524 +BPFP 1.3278 bits/point +EBPFP 1.3278 equivalent bits/point +MSE 369.285989 +---------------------- --------------------------------------------------------- +Time: 21.183s Load: 1.244s, Pack+Encode: 7.280s, Decode+Unpack: 12.659s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 369.2860 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00570-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00570-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst (62/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.265s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,774,840B, BPFP=1.1266 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,774,572B, BPFP=1.1264 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,249,840B, BPFP=1.4281 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,249,460B, BPFP=1.4278 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,510,012B, BPFP=1.5932 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,503,392B, BPFP=1.5890 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,602,928B, BPFP=1.0175 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,618,120B, BPFP=1.0271 +⌛️ [2/4] FRONTEND: Frontend time: 7.358s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.685s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00078516 4.39821978 + layer.9.1 0.00079184 8.57814976 + layer.19.0 3.22632161 19.90127991 + layer.19.1 3.22513146 15.12832482 + layer.29.0 0.10494786 22.92615829 + layer.29.1 0.10251782 37.75445087 + layer.39.0 10.88842496 1100.71213845 + layer.39.1 10.78217420 1139.64250894 + ------------------------------------------------------------------------------------- + TOTAL 3.54138686 293.63015385 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16283164 +BPFP 1.2920 bits/point +EBPFP 1.2920 equivalent bits/point +MSE 293.630154 +---------------------- --------------------------------------------------------- +Time: 21.308s Load: 1.265s, Pack+Encode: 7.358s, Decode+Unpack: 12.685s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 293.6302 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00758-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00758-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst (63/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.264s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,745,480B, BPFP=1.1079 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,709,004B, BPFP=1.0848 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,121,148B, BPFP=1.3464 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,104,692B, BPFP=1.3360 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,367,456B, BPFP=1.5027 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,367,332B, BPFP=1.5027 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,562,752B, BPFP=0.9920 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,559,436B, BPFP=0.9899 +⌛️ [2/4] FRONTEND: Frontend time: 7.406s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.655s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14561636 17.06030250 + layer.9.1 0.14552785 4.49382928 + layer.19.0 0.04069186 15.65367165 + layer.19.1 0.03840616 33.76096340 + layer.29.0 0.11346353 67.39947595 + layer.29.1 0.11182956 67.44721421 + layer.39.0 10.19697364 1130.25674358 + layer.39.1 10.11578978 1110.18979526 + ------------------------------------------------------------------------------------- + TOTAL 2.61353734 305.78274948 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15537300 +BPFP 1.2328 bits/point +EBPFP 1.2328 equivalent bits/point +MSE 305.782749 +---------------------- --------------------------------------------------------- +Time: 21.325s Load: 1.264s, Pack+Encode: 7.406s, Decode+Unpack: 12.655s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 305.7827 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00764-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00764-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst (64/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.241s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,785,284B, BPFP=1.1332 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,790,148B, BPFP=1.1363 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,280,688B, BPFP=1.4477 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,265,840B, BPFP=1.4382 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,541,920B, BPFP=1.6135 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,533,312B, BPFP=1.6080 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,636,960B, BPFP=1.0391 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,648,400B, BPFP=1.0463 +⌛️ [2/4] FRONTEND: Frontend time: 7.364s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.714s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14399445 8.44323560 + layer.9.1 0.14558028 12.55536160 + layer.19.0 0.03837104 29.02944731 + layer.19.1 0.04376782 71.34160708 + layer.29.0 0.11695251 66.74723757 + layer.29.1 0.13128335 95.66425292 + layer.39.0 11.28613757 1079.23350666 + layer.39.1 11.84408769 1105.18459539 + ------------------------------------------------------------------------------------- + TOTAL 2.96877184 308.52490552 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16482552 +BPFP 1.3078 bits/point +EBPFP 1.3078 equivalent bits/point +MSE 308.524906 +---------------------- --------------------------------------------------------- +Time: 21.318s Load: 1.241s, Pack+Encode: 7.364s, Decode+Unpack: 12.714s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 308.5249 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00845-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00845-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst (65/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.250s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,733,848B, BPFP=1.1006 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,725,724B, BPFP=1.0954 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,327,760B, BPFP=1.4775 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,332,444B, BPFP=1.4805 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,681,640B, BPFP=1.7022 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,676,232B, BPFP=1.6987 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,774,408B, BPFP=1.1263 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,799,452B, BPFP=1.1422 +⌛️ [2/4] FRONTEND: Frontend time: 7.382s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.732s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14502686 8.43653772 + layer.9.1 0.03259508 4.29602094 + layer.19.0 0.11326540 52.06160627 + layer.19.1 0.11324834 44.15580314 + layer.29.0 0.12250664 88.67999472 + layer.29.1 0.12058897 102.96008694 + layer.39.0 16.17915050 1394.99138772 + layer.39.1 21.66230805 1404.76308092 + ------------------------------------------------------------------------------------- + TOTAL 4.81108623 387.54306480 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17051508 +BPFP 1.3529 bits/point +EBPFP 1.3529 equivalent bits/point +MSE 387.543065 +---------------------- --------------------------------------------------------- +Time: 21.364s Load: 1.250s, Pack+Encode: 7.382s, Decode+Unpack: 12.732s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 387.5431 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/kitchen-rgb_00870-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/kitchen-rgb_00870-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst (66/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.261s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,734,444B, BPFP=1.1009 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,747,584B, BPFP=1.1093 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,204,880B, BPFP=1.3995 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,202,232B, BPFP=1.3979 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,461,284B, BPFP=1.5623 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,469,872B, BPFP=1.5678 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,587,328B, BPFP=1.0076 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,580,640B, BPFP=1.0033 +⌛️ [2/4] FRONTEND: Frontend time: 7.392s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.683s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66733019 4.42721610 + layer.9.1 2.66763138 8.45986350 + layer.19.0 3.22293078 10.45956200 + layer.19.1 3.22376992 15.55301176 + layer.29.0 4.27658332 20.33717780 + layer.29.1 4.27160529 29.68530884 + layer.39.0 7.81683598 1122.02453689 + layer.39.1 9.86231960 1147.21863828 + ------------------------------------------------------------------------------------- + TOTAL 4.75112581 294.77066440 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15988264 +BPFP 1.2686 bits/point +EBPFP 1.2686 equivalent bits/point +MSE 294.770664 +---------------------- --------------------------------------------------------- +Time: 21.337s Load: 1.261s, Pack+Encode: 7.392s, Decode+Unpack: 12.683s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 294.7707 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00153-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00153-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst (67/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.253s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,765,148B, BPFP=1.1204 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,779,032B, BPFP=1.1292 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,121,432B, BPFP=1.3466 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,133,708B, BPFP=1.3544 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,306,884B, BPFP=1.4643 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,314,768B, BPFP=1.4693 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,500,420B, BPFP=0.9524 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,503,904B, BPFP=0.9546 +⌛️ [2/4] FRONTEND: Frontend time: 7.370s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.642s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14549763 4.32891304 + layer.9.1 0.14520254 4.45930175 + layer.19.0 0.04746155 29.82637563 + layer.19.1 0.04383140 38.85198905 + layer.29.0 4.26247378 59.23650776 + layer.29.1 4.25497898 35.21493135 + layer.39.0 7.94138086 1182.90851479 + layer.39.1 7.86439079 1163.82328567 + ------------------------------------------------------------------------------------- + TOTAL 3.08815219 314.83122738 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15425296 +BPFP 1.2239 bits/point +EBPFP 1.2239 equivalent bits/point +MSE 314.831227 +---------------------- --------------------------------------------------------- +Time: 21.264s Load: 1.253s, Pack+Encode: 7.370s, Decode+Unpack: 12.642s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 314.8312 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00167-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00167-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst (68/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.249s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,715,208B, BPFP=1.0887 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,713,628B, BPFP=1.0877 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,055,396B, BPFP=1.3047 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,050,696B, BPFP=1.3017 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,243,904B, BPFP=1.4243 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,237,824B, BPFP=1.4205 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,494,708B, BPFP=0.9488 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,525,144B, BPFP=0.9681 +⌛️ [2/4] FRONTEND: Frontend time: 7.702s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.411s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.11283708 4.64016608 + layer.9.1 0.11300174 8.51743848 + layer.19.0 3.22718329 15.39837276 + layer.19.1 3.22892155 15.40857572 + layer.29.0 4.26448309 37.28134394 + layer.29.1 4.25758082 27.02322676 + layer.39.0 9.82393946 1286.98813780 + layer.39.1 9.78394007 1251.23578161 + ------------------------------------------------------------------------------------- + TOTAL 4.35148589 330.81163039 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15036508 +BPFP 1.1931 bits/point +EBPFP 1.1931 equivalent bits/point +MSE 330.811630 +---------------------- --------------------------------------------------------- +Time: 22.361s Load: 1.249s, Pack+Encode: 7.702s, Decode+Unpack: 13.411s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 330.8116 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_00208-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_00208-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst (69/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.236s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,882,964B, BPFP=1.1952 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,895,124B, BPFP=1.2029 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,297,720B, BPFP=1.4585 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,300,388B, BPFP=1.4602 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,489,924B, BPFP=1.5805 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,486,948B, BPFP=1.5786 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,672,460B, BPFP=1.0616 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,694,832B, BPFP=1.0758 +⌛️ [2/4] FRONTEND: Frontend time: 7.383s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.715s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14503861 25.45265529 + layer.9.1 0.14483112 29.59876706 + layer.19.0 0.11529889 30.02524019 + layer.19.1 0.11517203 40.02285099 + layer.29.0 0.11961639 70.50699748 + layer.29.1 0.11795276 85.09928502 + layer.39.0 83.84633978 1604.89015275 + layer.39.1 174.87768118 1584.03217420 + ------------------------------------------------------------------------------------- + TOTAL 32.43524135 433.70351537 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16720360 +BPFP 1.3267 bits/point +EBPFP 1.3267 equivalent bits/point +MSE 433.703515 +---------------------- --------------------------------------------------------- +Time: 21.335s Load: 1.236s, Pack+Encode: 7.383s, Decode+Unpack: 12.715s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 433.7035 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01201-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01201-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst (70/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.260s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,621,952B, BPFP=1.0295 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,635,836B, BPFP=1.0383 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,151,276B, BPFP=1.3655 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,135,704B, BPFP=1.3556 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,394,848B, BPFP=1.5201 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,403,264B, BPFP=1.5255 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,584,388B, BPFP=1.0057 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,608,492B, BPFP=1.0210 +⌛️ [2/4] FRONTEND: Frontend time: 7.435s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 13.195s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.81453913 12.56379362 + layer.9.1 0.14528001 4.47176094 + layer.19.0 3.26598681 24.31339881 + layer.19.1 0.04116655 15.70145688 + layer.29.0 4.28557138 131.14237691 + layer.29.1 4.28198282 151.78309839 + layer.39.0 74.89367180 1259.66582710 + layer.39.1 42.04871577 1253.59351641 + ------------------------------------------------------------------------------------- + TOTAL 16.47211428 356.65440363 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15535760 +BPFP 1.2327 bits/point +EBPFP 1.2327 equivalent bits/point +MSE 356.654404 +---------------------- --------------------------------------------------------- +Time: 21.890s Load: 1.260s, Pack+Encode: 7.435s, Decode+Unpack: 13.195s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 356.6544 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01296-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01296-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst (71/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.262s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,725,068B, BPFP=1.0950 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,740,596B, BPFP=1.1048 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,188,844B, BPFP=1.3894 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,195,520B, BPFP=1.3936 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,424,500B, BPFP=1.5390 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,429,340B, BPFP=1.5420 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,543,860B, BPFP=0.9800 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,533,664B, BPFP=0.9735 +⌛️ [2/4] FRONTEND: Frontend time: 7.329s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.638s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66710422 4.37526786 + layer.9.1 2.66812426 4.38439017 + layer.19.0 3.22059776 11.04330644 + layer.19.1 3.22546153 15.67977383 + layer.29.0 0.11226317 58.02787821 + layer.29.1 0.11257672 83.38785241 + layer.39.0 59.39237691 1227.61992200 + layer.39.1 37.52358222 1230.14299643 + ------------------------------------------------------------------------------------- + TOTAL 13.61526085 329.33267342 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15781392 +BPFP 1.2522 bits/point +EBPFP 1.2522 equivalent bits/point +MSE 329.332673 +---------------------- --------------------------------------------------------- +Time: 21.229s Load: 1.262s, Pack+Encode: 7.329s, Decode+Unpack: 12.638s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 329.3327 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/living_room-rgb_01313-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/living_room-rgb_01313-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst (72/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.250s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,734,716B, BPFP=1.1011 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,726,800B, BPFP=1.0961 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,222,836B, BPFP=1.4109 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,233,432B, BPFP=1.4177 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,501,100B, BPFP=1.5876 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,512,404B, BPFP=1.5947 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,546,544B, BPFP=0.9817 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,555,160B, BPFP=0.9871 +⌛️ [2/4] FRONTEND: Frontend time: 7.435s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.649s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14519123 4.31211471 + layer.9.1 0.14511500 4.32507503 + layer.19.0 0.03974548 15.54890620 + layer.19.1 0.03981401 10.67511476 + layer.29.0 4.26343511 58.58982979 + layer.29.1 4.25610090 48.38171311 + layer.39.0 7.90972018 1024.65030874 + layer.39.1 8.05601540 1058.29517387 + ------------------------------------------------------------------------------------- + TOTAL 3.10689216 278.09727953 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16032992 +BPFP 1.2721 bits/point +EBPFP 1.2721 equivalent bits/point +MSE 278.097280 +---------------------- --------------------------------------------------------- +Time: 21.334s Load: 1.250s, Pack+Encode: 7.435s, Decode+Unpack: 12.649s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 278.0973 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00032-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00032-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst (73/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.248s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,738,220B, BPFP=1.1033 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,737,536B, BPFP=1.1029 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,205,924B, BPFP=1.4002 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,196,080B, BPFP=1.3940 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,528,772B, BPFP=1.6051 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,513,532B, BPFP=1.5955 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,596,108B, BPFP=1.0131 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,576,144B, BPFP=1.0005 +⌛️ [2/4] FRONTEND: Frontend time: 7.368s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.698s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14586543 4.31851997 + layer.9.1 0.14572574 4.33515930 + layer.19.0 0.03953905 15.89633775 + layer.19.1 0.03760033 29.11376747 + layer.29.0 0.10448607 35.76962392 + layer.29.1 0.10697372 46.16305452 + layer.39.0 14.19073468 1134.41420214 + layer.39.1 8.92149669 1166.52323692 + ------------------------------------------------------------------------------------- + TOTAL 2.96155271 304.56673775 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16092316 +BPFP 1.2768 bits/point +EBPFP 1.2768 equivalent bits/point +MSE 304.566738 +---------------------- --------------------------------------------------------- +Time: 21.315s Load: 1.248s, Pack+Encode: 7.368s, Decode+Unpack: 12.698s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 304.5667 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00041-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00041-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst (74/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.240s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,757,708B, BPFP=1.1157 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,763,668B, BPFP=1.1195 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,285,060B, BPFP=1.4504 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,297,796B, BPFP=1.4585 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,550,624B, BPFP=1.6190 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,555,328B, BPFP=1.6220 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,645,168B, BPFP=1.0443 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,649,176B, BPFP=1.0468 +⌛️ [2/4] FRONTEND: Frontend time: 7.432s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.658s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00251883 12.55784728 + layer.9.1 0.14409062 12.36485416 + layer.19.0 0.12740102 48.75882048 + layer.19.1 0.12254588 93.78631784 + layer.29.0 4.25147928 43.65767692 + layer.29.1 4.25065697 48.40086224 + layer.39.0 9.21805114 1080.38844654 + layer.39.1 9.03214690 1073.80987975 + ------------------------------------------------------------------------------------- + TOTAL 3.39361133 301.71558815 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16504528 +BPFP 1.3095 bits/point +EBPFP 1.3095 equivalent bits/point +MSE 301.715588 +---------------------- --------------------------------------------------------- +Time: 21.330s Load: 1.240s, Pack+Encode: 7.432s, Decode+Unpack: 12.658s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 301.7156 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00042-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00042-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst (75/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.265s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,814,656B, BPFP=1.1519 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,821,728B, BPFP=1.1563 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,338,928B, BPFP=1.4846 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,329,796B, BPFP=1.4788 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,618,424B, BPFP=1.6620 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,619,688B, BPFP=1.6628 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,768,864B, BPFP=1.1228 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,763,720B, BPFP=1.1195 +⌛️ [2/4] FRONTEND: Frontend time: 7.419s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.706s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14550905 20.84715988 + layer.9.1 0.14590163 8.67944947 + layer.19.0 0.12839093 67.40652421 + layer.19.1 0.12422524 54.67880647 + layer.29.0 0.11695262 104.20900024 + layer.29.1 0.11389293 114.51451292 + layer.39.0 10.18180439 1225.64835879 + layer.39.1 10.42432323 1193.55362366 + ------------------------------------------------------------------------------------- + TOTAL 2.67262500 348.69217945 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17075804 +BPFP 1.3549 bits/point +EBPFP 1.3549 equivalent bits/point +MSE 348.692179 +---------------------- --------------------------------------------------------- +Time: 21.390s Load: 1.265s, Pack+Encode: 7.419s, Decode+Unpack: 12.706s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 348.6922 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00620-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00620-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst (76/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.242s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,779,044B, BPFP=1.1292 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,775,248B, BPFP=1.1268 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,269,544B, BPFP=1.4406 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,274,596B, BPFP=1.4438 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,604,908B, BPFP=1.6535 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,602,188B, BPFP=1.6517 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,689,440B, BPFP=1.0724 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,683,744B, BPFP=1.0688 +⌛️ [2/4] FRONTEND: Frontend time: 7.428s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.721s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14509775 12.86542797 + layer.9.1 0.14508723 12.80761877 + layer.19.0 0.11633494 38.63661592 + layer.19.1 0.11804005 57.33553888 + layer.29.0 0.15409572 164.50086326 + layer.29.1 0.14997486 96.38095141 + layer.39.0 9.23291952 1132.49878128 + layer.39.1 9.22304726 1155.11886578 + ------------------------------------------------------------------------------------- + TOTAL 2.41057467 333.76808291 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16678712 +BPFP 1.3234 bits/point +EBPFP 1.3234 equivalent bits/point +MSE 333.768083 +---------------------- --------------------------------------------------------- +Time: 21.390s Load: 1.242s, Pack+Encode: 7.428s, Decode+Unpack: 12.721s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 333.7681 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00636-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00636-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst (77/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.241s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,795,184B, BPFP=1.1395 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,793,620B, BPFP=1.1385 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,300,608B, BPFP=1.4603 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,303,624B, BPFP=1.4622 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,606,000B, BPFP=1.6542 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,609,864B, BPFP=1.6566 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,710,972B, BPFP=1.0860 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,719,056B, BPFP=1.0912 +⌛️ [2/4] FRONTEND: Frontend time: 7.402s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.662s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14497060 17.45972640 + layer.9.1 0.14492971 24.87914365 + layer.19.0 0.11929473 71.32386964 + layer.19.1 0.11869117 89.87110010 + layer.29.0 0.13715227 125.16504509 + layer.29.1 0.14278979 96.53747563 + layer.39.0 9.99110525 1200.11407215 + layer.39.1 10.01170034 1206.42866428 + ------------------------------------------------------------------------------------- + TOTAL 2.60132923 353.97238712 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16838928 +BPFP 1.3361 bits/point +EBPFP 1.3361 equivalent bits/point +MSE 353.972387 +---------------------- --------------------------------------------------------- +Time: 21.305s Load: 1.241s, Pack+Encode: 7.402s, Decode+Unpack: 12.662s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 353.9724 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office-rgb_00637-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office-rgb_00637-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst (78/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.242s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,713,040B, BPFP=1.0874 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,693,420B, BPFP=1.0749 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,254,536B, BPFP=1.4311 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,269,816B, BPFP=1.4408 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,573,796B, BPFP=1.6337 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,588,612B, BPFP=1.6431 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,759,356B, BPFP=1.1168 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,770,104B, BPFP=1.1236 +⌛️ [2/4] FRONTEND: Frontend time: 7.447s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.680s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14589856 4.76521622 + layer.9.1 0.03321603 8.55328026 + layer.19.0 0.11866178 48.86024842 + layer.19.1 0.11267978 75.55149090 + layer.29.0 0.10803594 78.30060225 + layer.29.1 0.10714094 93.16940201 + layer.39.0 11.58943751 1212.13950276 + layer.39.1 9.70079103 1227.13186545 + ------------------------------------------------------------------------------------- + TOTAL 2.73948269 343.55895103 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16622680 +BPFP 1.3189 bits/point +EBPFP 1.3189 equivalent bits/point +MSE 343.558951 +---------------------- --------------------------------------------------------- +Time: 21.369s Load: 1.242s, Pack+Encode: 7.447s, Decode+Unpack: 12.680s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 343.5590 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00410-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00410-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst (79/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.251s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,672,864B, BPFP=1.0619 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,685,252B, BPFP=1.0697 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,234,636B, BPFP=1.4184 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,232,104B, BPFP=1.4168 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,509,940B, BPFP=1.5932 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,518,660B, BPFP=1.5987 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,634,772B, BPFP=1.0377 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,620,740B, BPFP=1.0288 +⌛️ [2/4] FRONTEND: Frontend time: 7.404s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.656s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03308301 4.46749732 + layer.9.1 0.14566304 12.54470416 + layer.19.0 0.03810260 19.78797327 + layer.19.1 0.03780774 33.79733202 + layer.29.0 0.11592613 91.23355744 + layer.29.1 0.11717217 91.60389584 + layer.39.0 9.98032847 1120.24561261 + layer.39.1 9.70849498 1125.51218720 + ------------------------------------------------------------------------------------- + TOTAL 2.52207227 312.39909498 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16108968 +BPFP 1.2781 bits/point +EBPFP 1.2781 equivalent bits/point +MSE 312.399095 +---------------------- --------------------------------------------------------- +Time: 21.311s Load: 1.251s, Pack+Encode: 7.404s, Decode+Unpack: 12.656s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 312.3991 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00411-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00411-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst (80/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.244s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,689,424B, BPFP=1.0724 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,689,152B, BPFP=1.0722 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,192,744B, BPFP=1.3918 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,197,860B, BPFP=1.3951 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,379,224B, BPFP=1.5102 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,379,088B, BPFP=1.5101 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,484,932B, BPFP=0.9426 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,479,868B, BPFP=0.9393 +⌛️ [2/4] FRONTEND: Frontend time: 7.316s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.661s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14555504 8.62321572 + layer.9.1 0.14557384 8.46692192 + layer.19.0 0.03995539 43.20040319 + layer.19.1 0.04542811 33.97814430 + layer.29.0 0.12033866 76.63518646 + layer.29.1 0.13252172 120.19087179 + layer.39.0 10.37566776 1077.29444264 + layer.39.1 9.84188447 987.27006825 + ------------------------------------------------------------------------------------- + TOTAL 2.60586562 294.45740678 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15492292 +BPFP 1.2292 bits/point +EBPFP 1.2292 equivalent bits/point +MSE 294.457407 +---------------------- --------------------------------------------------------- +Time: 21.220s Load: 1.244s, Pack+Encode: 7.316s, Decode+Unpack: 12.661s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 294.4574 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00412-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00412-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst (81/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.240s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,778,132B, BPFP=1.1287 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,776,092B, BPFP=1.1274 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,333,012B, BPFP=1.4809 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,333,448B, BPFP=1.4812 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,642,392B, BPFP=1.6773 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,644,216B, BPFP=1.6784 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,848,216B, BPFP=1.1732 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,859,060B, BPFP=1.1800 +⌛️ [2/4] FRONTEND: Frontend time: 7.406s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.764s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14470203 12.56712986 + layer.9.1 0.14481130 12.56592003 + layer.19.0 0.11257574 53.37714292 + layer.19.1 0.11422884 56.99916213 + layer.29.0 0.10456927 64.53881114 + layer.29.1 0.10551051 112.03257028 + layer.39.0 10.36536069 1236.78794280 + layer.39.1 11.81531702 1252.41233344 + ------------------------------------------------------------------------------------- + TOTAL 2.86338443 350.16012658 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17214568 +BPFP 1.3659 bits/point +EBPFP 1.3659 equivalent bits/point +MSE 350.160127 +---------------------- --------------------------------------------------------- +Time: 21.409s Load: 1.240s, Pack+Encode: 7.406s, Decode+Unpack: 12.764s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 350.1601 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/office_kitchen-rgb_00413-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/office_kitchen-rgb_00413-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst (82/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.245s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,755,804B, BPFP=1.1145 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,768,328B, BPFP=1.1224 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,306,356B, BPFP=1.4640 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,294,376B, BPFP=1.4564 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,492,740B, BPFP=1.5823 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,484,040B, BPFP=1.5767 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,592,904B, BPFP=1.0111 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,591,380B, BPFP=1.0101 +⌛️ [2/4] FRONTEND: Frontend time: 7.420s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.644s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14530003 8.41748190 + layer.9.1 0.14546206 12.54993576 + layer.19.0 0.11891763 47.39133084 + layer.19.1 0.11677460 52.64977047 + layer.29.0 4.29725807 82.73297855 + layer.29.1 4.29692800 53.95229729 + layer.39.0 11.61914761 1152.16777706 + layer.39.1 11.22064282 1127.09717257 + ------------------------------------------------------------------------------------- + TOTAL 3.99505385 317.11984305 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16285928 +BPFP 1.2922 bits/point +EBPFP 1.2922 equivalent bits/point +MSE 317.119843 +---------------------- --------------------------------------------------------- +Time: 21.309s Load: 1.245s, Pack+Encode: 7.420s, Decode+Unpack: 12.644s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 317.1198 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00433-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00433-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst (83/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.299s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,658,024B, BPFP=1.0524 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,657,132B, BPFP=1.0519 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,127,632B, BPFP=1.3505 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,140,644B, BPFP=1.3588 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,352,228B, BPFP=1.4931 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,355,984B, BPFP=1.4955 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,606,828B, BPFP=1.0199 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,614,404B, BPFP=1.0247 +⌛️ [2/4] FRONTEND: Frontend time: 7.388s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.661s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.67106792 4.51059334 + layer.9.1 2.67195307 4.37351342 + layer.19.0 0.08237472 34.33779859 + layer.19.1 0.08192194 29.30520292 + layer.29.0 0.11152953 134.77057605 + layer.29.1 0.11703055 137.91242485 + layer.39.0 163.01811830 1431.94962626 + layer.39.1 58.15221299 1439.54939877 + ------------------------------------------------------------------------------------- + TOTAL 28.36327613 402.08864177 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15512876 +BPFP 1.2308 bits/point +EBPFP 1.2308 equivalent bits/point +MSE 402.088642 +---------------------- --------------------------------------------------------- +Time: 21.347s Load: 1.299s, Pack+Encode: 7.388s, Decode+Unpack: 12.661s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 402.0886 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00443-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00443-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst (84/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.241s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,816,796B, BPFP=1.1532 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,814,700B, BPFP=1.1519 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,230,132B, BPFP=1.4156 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,226,080B, BPFP=1.4130 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,472,012B, BPFP=1.5691 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,476,428B, BPFP=1.5719 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,635,372B, BPFP=1.0381 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,600,336B, BPFP=1.0158 +⌛️ [2/4] FRONTEND: Frontend time: 7.408s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.711s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14620710 20.87513457 + layer.9.1 0.14642976 25.39810743 + layer.19.0 0.11726453 52.75999858 + layer.19.1 0.11958517 34.22399557 + layer.29.0 0.10693079 34.66269144 + layer.29.1 0.10826971 44.13805553 + layer.39.0 43.01306569 1288.60350991 + layer.39.1 17.12450997 1351.88844654 + ------------------------------------------------------------------------------------- + TOTAL 7.61028284 356.56874245 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16271856 +BPFP 1.2911 bits/point +EBPFP 1.2911 equivalent bits/point +MSE 356.568742 +---------------------- --------------------------------------------------------- +Time: 21.360s Load: 1.241s, Pack+Encode: 7.408s, Decode+Unpack: 12.711s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 356.5687 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00444-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00444-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst (85/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.247s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,692,680B, BPFP=1.0744 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,700,132B, BPFP=1.0792 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,128,056B, BPFP=1.3508 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,131,956B, BPFP=1.3533 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,370,000B, BPFP=1.5044 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,372,076B, BPFP=1.5057 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,526,080B, BPFP=0.9687 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,558,668B, BPFP=0.9894 +⌛️ [2/4] FRONTEND: Frontend time: 7.378s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.653s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.03173930 4.35539544 + layer.9.1 0.03345565 4.42492846 + layer.19.0 3.26068347 24.54671758 + layer.19.1 3.26087326 24.74740768 + layer.29.0 4.24610771 29.51242586 + layer.29.1 4.24089229 29.29130698 + layer.39.0 8.81319124 1068.31897953 + layer.39.1 8.71779153 1099.69345141 + ------------------------------------------------------------------------------------- + TOTAL 4.07559181 285.61132662 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 15479648 +BPFP 1.2282 bits/point +EBPFP 1.2282 equivalent bits/point +MSE 285.611327 +---------------------- --------------------------------------------------------- +Time: 21.278s Load: 1.247s, Pack+Encode: 7.378s, Decode+Unpack: 12.653s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 285.6113 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00445-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00445-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst (86/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.249s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,731,976B, BPFP=1.0994 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,728,556B, BPFP=1.0972 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,247,576B, BPFP=1.4266 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,247,384B, BPFP=1.4265 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,569,192B, BPFP=1.6308 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,585,872B, BPFP=1.6414 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,644,916B, BPFP=1.0441 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,672,640B, BPFP=1.0617 +⌛️ [2/4] FRONTEND: Frontend time: 7.237s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.602s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00078599 4.37353214 + layer.9.1 0.00079117 16.56757800 + layer.19.0 0.00795310 24.98992779 + layer.19.1 0.00811505 20.18150796 + layer.29.0 4.25797468 30.51122238 + layer.29.1 4.25504309 68.83226865 + layer.39.0 81.06806549 1233.38137797 + layer.39.1 44.82015254 1185.45352616 + ------------------------------------------------------------------------------------- + TOTAL 16.80236014 323.03636763 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16428112 +BPFP 1.3035 bits/point +EBPFP 1.3035 equivalent bits/point +MSE 323.036368 +---------------------- --------------------------------------------------------- +Time: 21.089s Load: 1.249s, Pack+Encode: 7.237s, Decode+Unpack: 12.602s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 323.0364 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00446-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00446-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst (87/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.260s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,718,568B, BPFP=1.0909 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,724,448B, BPFP=1.0946 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,269,808B, BPFP=1.4408 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,276,736B, BPFP=1.4452 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,604,388B, BPFP=1.6531 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,613,216B, BPFP=1.6587 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,677,084B, BPFP=1.0645 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,672,440B, BPFP=1.0616 +⌛️ [2/4] FRONTEND: Frontend time: 7.426s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.702s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00082351 4.36896956 + layer.9.1 0.02968625 8.43630604 + layer.19.0 0.00841222 29.23052080 + layer.19.1 0.03743129 20.29354002 + layer.29.0 4.28408194 48.12985964 + layer.29.1 4.28564945 43.62060753 + layer.39.0 8.35370986 1053.95466363 + layer.39.1 8.52557915 1057.85667858 + ------------------------------------------------------------------------------------- + TOTAL 3.19067171 283.23639323 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16556688 +BPFP 1.3137 bits/point +EBPFP 1.3137 equivalent bits/point +MSE 283.236393 +---------------------- --------------------------------------------------------- +Time: 21.388s Load: 1.260s, Pack+Encode: 7.426s, Decode+Unpack: 12.702s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 283.2364 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/playroom-rgb_00447-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/playroom-rgb_00447-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst (88/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.250s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,727,156B, BPFP=1.0963 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,732,564B, BPFP=1.0997 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,223,164B, BPFP=1.4112 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,234,180B, BPFP=1.4181 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,486,280B, BPFP=1.5782 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,490,180B, BPFP=1.5806 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,613,172B, BPFP=1.0240 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,625,508B, BPFP=1.0318 +⌛️ [2/4] FRONTEND: Frontend time: 7.245s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.605s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14529990 4.32054196 + layer.9.1 0.14524076 4.31855234 + layer.19.0 0.03780325 33.94206004 + layer.19.1 0.03783790 25.06633897 + layer.29.0 4.32098184 127.16603022 + layer.29.1 4.32100596 93.40083685 + layer.39.0 9.32673680 1224.20076373 + layer.39.1 9.31823369 1269.37455314 + ------------------------------------------------------------------------------------- + TOTAL 3.45664251 347.72370966 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16132204 +BPFP 1.2800 bits/point +EBPFP 1.2800 equivalent bits/point +MSE 347.723710 +---------------------- --------------------------------------------------------- +Time: 21.100s Load: 1.250s, Pack+Encode: 7.245s, Decode+Unpack: 12.605s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 347.7237 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00461-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00461-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst (89/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.246s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,739,444B, BPFP=1.1041 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,759,792B, BPFP=1.1170 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,251,868B, BPFP=1.4294 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,244,636B, BPFP=1.4248 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,526,880B, BPFP=1.6039 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,525,444B, BPFP=1.6030 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,657,008B, BPFP=1.0518 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,657,456B, BPFP=1.0521 +⌛️ [2/4] FRONTEND: Frontend time: 7.403s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.680s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14497286 4.30538128 + layer.9.1 0.14497296 8.50744942 + layer.19.0 0.03962668 62.39112264 + layer.19.1 0.11751332 80.21643443 + layer.29.0 0.14529291 112.35087951 + layer.29.1 0.16241527 145.11167533 + layer.39.0 11.40179406 1265.69231394 + layer.39.1 13.03458244 1255.76072473 + ------------------------------------------------------------------------------------- + TOTAL 3.14889631 366.79199766 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16362528 +BPFP 1.2983 bits/point +EBPFP 1.2983 equivalent bits/point +MSE 366.791998 +---------------------- --------------------------------------------------------- +Time: 21.328s Load: 1.246s, Pack+Encode: 7.403s, Decode+Unpack: 12.680s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 366.7920 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00462-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00462-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst (90/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.243s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,834,244B, BPFP=1.1643 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,824,580B, BPFP=1.1582 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,322,212B, BPFP=1.4740 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,324,096B, BPFP=1.4752 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,609,268B, BPFP=1.6562 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,620,804B, BPFP=1.6636 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,720,808B, BPFP=1.0923 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,720,180B, BPFP=1.0919 +⌛️ [2/4] FRONTEND: Frontend time: 7.249s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.636s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14513344 12.44439338 + layer.9.1 0.03283094 4.31482794 + layer.19.0 0.11544709 38.98884872 + layer.19.1 0.11326018 57.61467135 + layer.29.0 0.14483232 154.71257515 + layer.29.1 0.14672551 145.26186220 + layer.39.0 10.02784076 1157.18817030 + layer.39.1 15.62606130 1209.96595710 + ------------------------------------------------------------------------------------- + TOTAL 3.29401644 347.56141327 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16976192 +BPFP 1.3470 bits/point +EBPFP 1.3470 equivalent bits/point +MSE 347.561413 +---------------------- --------------------------------------------------------- +Time: 21.128s Load: 1.243s, Pack+Encode: 7.249s, Decode+Unpack: 12.636s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 347.5614 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00463-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00463-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst (91/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.250s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,874,860B, BPFP=1.1901 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,883,232B, BPFP=1.1954 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,374,572B, BPFP=1.5073 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,379,220B, BPFP=1.5102 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,645,888B, BPFP=1.6795 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,659,292B, BPFP=1.6880 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,746,328B, BPFP=1.1085 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,770,560B, BPFP=1.1239 +⌛️ [2/4] FRONTEND: Frontend time: 7.465s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.749s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14492303 8.58071732 + layer.9.1 0.14484742 8.37423132 + layer.19.0 0.11740684 71.71286460 + layer.19.1 0.11489933 39.39322240 + layer.29.0 0.12072669 108.86040583 + layer.29.1 0.12118037 74.91024638 + layer.39.0 10.74778980 1193.72107572 + layer.39.1 11.83662176 1199.91070848 + ------------------------------------------------------------------------------------- + TOTAL 2.91854940 338.18293401 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17333952 +BPFP 1.3753 bits/point +EBPFP 1.3753 equivalent bits/point +MSE 338.182934 +---------------------- --------------------------------------------------------- +Time: 21.464s Load: 1.250s, Pack+Encode: 7.465s, Decode+Unpack: 12.749s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 338.1829 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00464-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00464-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst (92/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.245s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,922,852B, BPFP=1.2205 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,923,416B, BPFP=1.2209 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,400,916B, BPFP=1.5240 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,392,380B, BPFP=1.5186 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,684,564B, BPFP=1.7040 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,672,144B, BPFP=1.6961 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,789,076B, BPFP=1.1356 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,771,784B, BPFP=1.1246 +⌛️ [2/4] FRONTEND: Frontend time: 7.255s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.696s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14465000 8.83476067 + layer.9.1 0.14489275 8.74420664 + layer.19.0 0.11978787 104.16943248 + layer.19.1 0.12819003 104.58871263 + layer.29.0 0.12519148 113.76754956 + layer.29.1 0.13018718 108.94167411 + layer.39.0 10.77894586 1251.94572636 + layer.39.1 10.25834823 1194.32539812 + ------------------------------------------------------------------------------------- + TOTAL 2.72877418 361.91468257 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17557132 +BPFP 1.3930 bits/point +EBPFP 1.3930 equivalent bits/point +MSE 361.914683 +---------------------- --------------------------------------------------------- +Time: 21.195s Load: 1.245s, Pack+Encode: 7.255s, Decode+Unpack: 12.696s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 361.9147 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/reception_room-rgb_00465-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/reception_room-rgb_00465-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst (93/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.248s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,767,876B, BPFP=1.1222 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,772,516B, BPFP=1.1251 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,209,124B, BPFP=1.4022 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,222,924B, BPFP=1.4110 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,444,996B, BPFP=1.5520 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,460,596B, BPFP=1.5619 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,627,784B, BPFP=1.0332 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,643,516B, BPFP=1.0432 +⌛️ [2/4] FRONTEND: Frontend time: 7.232s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.623s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14553812 4.34753406 + layer.9.1 0.14559401 8.56206837 + layer.19.0 0.04492324 43.99113889 + layer.19.1 0.04213941 48.65074545 + layer.29.0 4.25320263 34.84802567 + layer.29.1 4.25391672 53.98905184 + layer.39.0 8.72311137 1147.86927202 + layer.39.1 8.87262096 1170.37292818 + ------------------------------------------------------------------------------------- + TOTAL 3.31013081 314.07884556 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16149332 +BPFP 1.2813 bits/point +EBPFP 1.2813 equivalent bits/point +MSE 314.078846 +---------------------- --------------------------------------------------------- +Time: 21.103s Load: 1.248s, Pack+Encode: 7.232s, Decode+Unpack: 12.623s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 314.0788 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00471-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00471-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst (94/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.242s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,798,152B, BPFP=1.1414 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,794,096B, BPFP=1.1388 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,271,520B, BPFP=1.4418 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,277,504B, BPFP=1.4456 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,552,656B, BPFP=1.6203 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,554,768B, BPFP=1.6216 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,643,044B, BPFP=1.0429 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,659,540B, BPFP=1.0534 +⌛️ [2/4] FRONTEND: Frontend time: 7.264s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.585s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14518122 12.59417528 + layer.9.1 0.14529820 4.58789570 + layer.19.0 0.11833418 71.59023095 + layer.19.1 0.12038008 62.35634851 + layer.29.0 4.31360161 194.31231719 + layer.29.1 4.31792870 169.16723879 + layer.39.0 9.40764201 1200.98781280 + layer.39.1 11.30764416 1198.15599610 + ------------------------------------------------------------------------------------- + TOTAL 3.73450127 364.21900192 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16551280 +BPFP 1.3132 bits/point +EBPFP 1.3132 equivalent bits/point +MSE 364.219002 +---------------------- --------------------------------------------------------- +Time: 21.092s Load: 1.242s, Pack+Encode: 7.264s, Decode+Unpack: 12.585s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 364.2190 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00472-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00472-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst (95/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.242s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,910,380B, BPFP=1.2126 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,906,324B, BPFP=1.2100 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,395,672B, BPFP=1.5207 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,397,304B, BPFP=1.5217 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,673,452B, BPFP=1.6970 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,680,876B, BPFP=1.7017 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,725,920B, BPFP=1.0955 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,706,296B, BPFP=1.0831 +⌛️ [2/4] FRONTEND: Frontend time: 7.325s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.684s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00315138 21.19692096 + layer.9.1 0.00505826 45.94053156 + layer.19.0 0.09147678 62.73672611 + layer.19.1 0.09143778 62.09115514 + layer.29.0 0.11015094 83.81665380 + layer.29.1 0.11338039 92.76533555 + layer.39.0 9.14784464 1069.87845304 + layer.39.1 8.98944348 1065.76088723 + ------------------------------------------------------------------------------------- + TOTAL 2.31899296 313.02333292 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17396224 +BPFP 1.3803 bits/point +EBPFP 1.3803 equivalent bits/point +MSE 313.023333 +---------------------- --------------------------------------------------------- +Time: 21.252s Load: 1.242s, Pack+Encode: 7.325s, Decode+Unpack: 12.684s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 313.0233 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00475-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00475-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst (96/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.246s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,920,792B, BPFP=1.2192 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,924,156B, BPFP=1.2214 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,459,556B, BPFP=1.5612 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,452,396B, BPFP=1.5567 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,768,432B, BPFP=1.7573 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,759,728B, BPFP=1.7517 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,860,952B, BPFP=1.1812 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,832,872B, BPFP=1.1634 +⌛️ [2/4] FRONTEND: Frontend time: 7.438s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.793s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14591913 21.89487275 + layer.9.1 0.03347605 21.43611371 + layer.19.0 0.12173996 66.64527645 + layer.19.1 0.12099332 99.18335635 + layer.29.0 0.11078974 65.09959478 + layer.29.1 0.11776269 59.77051511 + layer.39.0 10.17800795 1118.06548586 + layer.39.1 9.88744998 1126.53883653 + ------------------------------------------------------------------------------------- + TOTAL 2.58951735 322.32925644 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17978884 +BPFP 1.4265 bits/point +EBPFP 1.4265 equivalent bits/point +MSE 322.329256 +---------------------- --------------------------------------------------------- +Time: 21.476s Load: 1.246s, Pack+Encode: 7.438s, Decode+Unpack: 12.793s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 322.3293 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00476-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00476-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst (97/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.250s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,726,604B, BPFP=1.0960 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,730,996B, BPFP=1.0987 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,258,536B, BPFP=1.4336 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,266,856B, BPFP=1.4389 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,616,328B, BPFP=1.6607 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,624,496B, BPFP=1.6659 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,573,564B, BPFP=0.9988 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,602,480B, BPFP=1.0172 +⌛️ [2/4] FRONTEND: Frontend time: 7.773s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.643s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 2.66405266 4.46067186 + layer.9.1 2.66543197 4.58401294 + layer.19.0 3.22131407 10.73510369 + layer.19.1 3.22426883 15.48629956 + layer.29.0 4.27224607 50.13760867 + layer.29.1 4.27784520 44.85840510 + layer.39.0 8.94937744 1036.96246344 + layer.39.1 8.82170070 1009.00333117 + ------------------------------------------------------------------------------------- + TOTAL 4.76202962 272.02848705 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16399860 +BPFP 1.3012 bits/point +EBPFP 1.3012 equivalent bits/point +MSE 272.028487 +---------------------- --------------------------------------------------------- +Time: 21.666s Load: 1.250s, Pack+Encode: 7.773s, Decode+Unpack: 12.643s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 272.0285 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00643-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00643-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst (98/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.249s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,795,416B, BPFP=1.1396 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,793,444B, BPFP=1.1384 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,318,044B, BPFP=1.4714 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,314,476B, BPFP=1.4691 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,647,120B, BPFP=1.6803 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,634,632B, BPFP=1.6723 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,706,376B, BPFP=1.0831 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,690,024B, BPFP=1.0727 +⌛️ [2/4] FRONTEND: Frontend time: 7.282s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.670s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.00104424 8.73521731 + layer.9.1 0.00091568 4.53917199 + layer.19.0 0.08171424 43.01876320 + layer.19.1 0.08373584 52.31667919 + layer.29.0 4.26071267 34.09279280 + layer.29.1 4.26438533 34.29127143 + layer.39.0 8.39843369 1055.36488463 + layer.39.1 8.51949380 994.11147221 + ------------------------------------------------------------------------------------- + TOTAL 3.20130444 278.30878160 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16899532 +BPFP 1.3409 bits/point +EBPFP 1.3409 equivalent bits/point +MSE 278.308782 +---------------------- --------------------------------------------------------- +Time: 21.201s Load: 1.249s, Pack+Encode: 7.282s, Decode+Unpack: 12.670s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 278.3088 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study-rgb_00644-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study-rgb_00644-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst (99/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.252s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,883,296B, BPFP=1.1954 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,877,784B, BPFP=1.1919 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,367,784B, BPFP=1.5030 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,371,952B, BPFP=1.5056 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,680,752B, BPFP=1.7016 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,681,972B, BPFP=1.7024 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,770,888B, BPFP=1.1241 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,759,540B, BPFP=1.1169 +⌛️ [2/4] FRONTEND: Frontend time: 7.440s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.711s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14622840 8.35651736 + layer.9.1 0.03344178 13.20757333 + layer.19.0 0.12675888 108.29523481 + layer.19.1 0.12382618 107.42161602 + layer.29.0 0.12223263 97.50207182 + layer.29.1 0.12797405 68.35935469 + layer.39.0 10.69978368 1149.14673383 + layer.39.1 8.63538768 1195.77266818 + ------------------------------------------------------------------------------------- + TOTAL 2.50195416 343.50772125 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 17393968 +BPFP 1.3801 bits/point +EBPFP 1.3801 equivalent bits/point +MSE 343.507721 +---------------------- --------------------------------------------------------- +Time: 21.403s Load: 1.252s, Pack+Encode: 7.440s, Decode+Unpack: 12.711s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 343.5077 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00272-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00272-wflip.zst + + 💪 Processing: ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst (100/100) + +[1/4] FRONTEND: Loading features from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst... + +Original data structure: +root: [dict] with 4 keys + key['encoder.backbone.blocks.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['encoder.backbone.blocks.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + +Parsed features: +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ +⌛️ [1/4] FRONTEND: Load time: 1.245s + +------------------------------------------------------------ +DINOv3 Features Summary +------------------------------------------------------------ +Number of layers: 8 +Layer names: ['layer.9', 'layer.9', 'layer.19', 'layer.19', 'layer.29', 'layer.29', 'layer.39', 'layer.39'] +Data type: torch.float32 + layer.9.0: torch.Size([3077, 4096]), torch.float32 + layer.9.1: torch.Size([3077, 4096]), torch.float32 + layer.19.0: torch.Size([3077, 4096]), torch.float32 + layer.19.1: torch.Size([3077, 4096]), torch.float32 + layer.29.0: torch.Size([3077, 4096]), torch.float32 + layer.29.1: torch.Size([3077, 4096]), torch.float32 + layer.39.0: torch.Size([3077, 4096]), torch.float32 + layer.39.1: torch.Size([3077, 4096]), torch.float32 +------------------------------------------------------------ + +[2/4] FRONTEND: Pack + Encode (strategy: individual)... + IndividualPacker: + layer.9.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.0 + layer.9.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.9.1 + layer.19.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.0 + layer.19.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.19.1 + layer.29.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.0 + layer.29.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.29.1 + layer.39.0: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.0 + layer.39.1: torch.Size([3077, 4096]) -> torch.Size([1, 1, 3077, 4096]) + From layer.39.1 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + layer.9.0: 1,829,076B, BPFP=1.1610 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + layer.9.1: 1,826,720B, BPFP=1.1595 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + layer.19.0: 2,308,420B, BPFP=1.4653 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + layer.19.1: 2,305,496B, BPFP=1.4634 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + layer.29.0: 2,532,760B, BPFP=1.6077 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + layer.29.1: 2,530,872B, BPFP=1.6065 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + layer.39.0: 1,648,288B, BPFP=1.0463 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + layer.39.1: 1,659,908B, BPFP=1.0536 +⌛️ [2/4] FRONTEND: Frontend time: 7.415s (Pack+Encode) + +[3/4] BACKEND: Decode + Unpack... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) +⌛️ [3/4] BACKEND: Backend time: 12.722s + +[4/4] METRICS: Computing MSE Breakdown... + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.0 + Using per-key quantization points (layer.9: torch.Size([256])) for layer.9.1 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.0 + Using per-key quantization points (layer.19: torch.Size([256])) for layer.19.1 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.0 + Using per-key quantization points (layer.29: torch.Size([256])) for layer.29.1 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.0 + Using per-key quantization points (layer.39: torch.Size([256])) for layer.39.1 + IndividualUnPacker: + layer.9.0: torch.Size([1, 3077, 4096]) + layer.9.1: torch.Size([1, 3077, 4096]) + layer.19.0: torch.Size([1, 3077, 4096]) + layer.19.1: torch.Size([1, 3077, 4096]) + layer.29.0: torch.Size([1, 3077, 4096]) + layer.29.1: torch.Size([1, 3077, 4096]) + layer.39.0: torch.Size([1, 3077, 4096]) + layer.39.1: torch.Size([1, 3077, 4096]) + Per-key MSE Breakdown (Quant=Truncation+Quantization, Codec=Compression): + Key Quant-MSE Total-MSE + ------------------------------------------------------------------------------------- + layer.9.0 0.14492994 4.57359320 + layer.9.1 0.14498602 25.02116256 + layer.19.0 0.12957112 47.94749350 + layer.19.1 0.13054295 61.97921575 + layer.29.0 0.16610158 197.90674764 + layer.29.1 0.14872770 159.53390071 + layer.39.0 16.52878844 1261.45084498 + layer.39.1 24.55764797 1268.31946701 + ------------------------------------------------------------------------------------- + TOTAL 5.24391197 378.34155317 + (elements=100,827,136) +---------------------- --------------------------------------------------------- +SAMPLE-WISE STATISTICS +---------------------- --------------------------------------------------------- +Handler dinov3-total +Strategy individual +Architecture hyperprior-featurecoding +---------------------- --------------------------------------------------------- +Total Elements 100827136 +Total Bytes 16641540 +BPFP 1.3204 bits/point +EBPFP 1.3204 equivalent bits/point +MSE 378.341553 +---------------------- --------------------------------------------------------- +Time: 21.382s Load: 1.245s, Pack+Encode: 7.415s, Decode+Unpack: 12.722s +---------------------- --------------------------------------------------------- +Restored Feature Format: [dict] with 4 keys + key['layer.9']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.19']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.29']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu + key['layer.39']: [Tensor] shape=torch.Size([2, 3077, 4096]), dtype=torch.float32, device=cpu +💾 Converting with 378.3416 MSE: + from ../datasets/FeatureCoding-DINOv3/DINOv3Dep-NYUDv2Test-4lvl-100Features/study_room-rgb_00278-wflip.zst + to output-fixed/dinov3-total/lambda0.02/hyperprior-featurecoding-8bit-individual/nyudv2_test/study_room-rgb_00278-wflip.zst +------------------------ ---------------------------- +TOTAL PROCESSING SUMMARY +------------------------ ---------------------------- +Total files 100 +Avg BPFP 1.3061 bits/point +Avg EBPFP 1.3061 equivalent bits/point +Avg MSE 335.304397 +Avg Time 21.635s +------------------------ ----------------------------