Add ReMoMask-2 inference assets

README.md

@@ -0,0 +1,40 @@
+---
+language:
+- en
+tags:
+- text-to-motion
+- motion-generation
+- retrieval-augmented-generation
+---
+
+# ReMoMask-2: Latent Retrieval-Augmented Masked Motion Generation
+
+> Yiran Wang<sup>*</sup>, Zeyu Zhang<sup>*†</sup>, Philip Torr, and Hao Tang<sup>‡</sup>
+>
+> <sup>*</sup>Equal contribution. <sup>†</sup>Project lead. <sup>‡</sup>Corresponding author.
+>
+> **arXiv:** Coming soon · **Project Page:** Coming soon
+
+## Inference assets
+
+| Files | Contents |
+| --- | --- |
+| `net_best_fid_ep0309.tar` | Masked motion generation model |
+| `vq.tar` | RVQ model |
+| `query_projector.pt` | Text-to-motion query projector |
+| `model_opt.txt`, `vq_opt.txt` | Model configurations |
+| `mean.npy`, `std.npy` | Motion normalization |
+| `encoded_motions.npy`, `encoded_texts_clip.npy` | Retrieval features |
+| `motion_ids.npy`, `all_captions.npy` | Retrieval metadata |
+
+## Download
+
+```python
+from huggingface_hub import snapshot_download
+
+snapshot_download(
+    "yiranranranra/ReMoMask-2",
+    local_dir="remomask2",
+    allow_patterns=["*.tar", "*.pt", "*.txt", "*.npy"],
+)
+```

all_captions.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:85793c088f56a41d867ae0605ffeb56a6d0668f88928169dea7346bf4a4c831d
+size 240432008

encoded_motions.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2038efb74aec708418647545917f8d911f0432b98333c0ff1bdf8facb894cfcb
+size 274473088

encoded_texts_clip.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3bcb7236ae7debbf87a861c22cf967f7f7445ce07004863c6c9a640ecc5b93d8
+size 137236608

mean.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d70b13c5ff0e2d7d8510fd3cc70859c526ebe048765ab33b4f95f98f5ecbfcbe
+size 2232

model_opt.txt

@@ -0,0 +1,44 @@
+------------ Options -------------
+attnj: True
+attnt: True
+batch_size: 64
+checkpoints_dir: ./logs
+cond_drop_prob: 0.1
+dataset_name: humanml3d
+dropout: 0.2
+eval_every_e: 10
+ff_size: 1024
+force_mask: False
+gamma: 0.1
+gpu_id: -1
+gumbel_sample: False
+is_continue: False
+is_train: True
+latent_dim: 512
+local_rank: 0
+log_every: 50
+lr: 0.0002
+max_epoch: 2000
+max_motion_length: 196
+milestones: [1000000]
+mtrans_warmstart_ckpt: None
+mtrans_warmstart_epoch: net_best_fid_ep0486
+n_heads: 8
+n_layers: 8
+name: v2_mtrans_vgate
+projector_path: logs/query_projector_repair_ep/best_projector.pt
+retr_vgate: True
+retrieval_dim: None
+rt_in_value: True
+save_latest: 500
+seed: 3407
+share_weight: False
+single_uncond_dice: False
+train_split: train.txt
+unit_length: 4
+use_ze_retrieval: True
+val_split: val.txt
+vq_name: pretrain_vq
+warm_up_iter: 2000
+ze_database_path: database_ze
+-------------- End ----------------

motion_ids.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b8f7f45dfbaffb0ff83c5b6961a5332a2d714118d3c4156252c0f23f63e48c6
+size 2412488

net_best_fid_ep0309.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7412fab178c5e863f292fdb0b39d322f86656540f617ed03f4eb8f1bf7b7d422
+size 344691218

query_projector.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1aaab3bc7254d967a9a1bb6719208eb8afc20099fa083e5e2cdad93c10595d43
+size 6302216

std.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fcd6fdfe3f068c23fabfe6587119fb8799a17d405e28508dc3d1a8f2d58a0119
+size 2232

vq.tar

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b186fd346e8bcb84cdbfc2f2c78a9ea3f67c92389050c178325897de6b5de767
+size 467921538

vq_opt.txt

@@ -0,0 +1,43 @@
+------------ Options -------------
+batch_size: 256
+checkpoints_dir: ./logs
+code_dim1d: 512
+code_dim2d: 1024
+commit: 0.02
+dataset_name: humanml3d
+depth: 3
+dilation_growth_rate: 3
+down_t: 2
+eval_every_e: 1
+ext: default
+feat_bias: 5
+gamma: 0.1
+gpu_id: 2
+is_continue: False
+is_train: True
+log_every: 10
+loss_vel: 0.5
+lr: 0.0002
+max_epoch: 50
+milestones: [150000, 250000]
+mu: 0.99
+name: pretrain_vq
+nb_code1d: 512
+nb_code2d: 256
+num_quantizers: 6
+output_emb_width: 512
+quantize_dropout_prob: 0.2
+recons_loss: l1_smooth
+save_every_e: 2
+save_latest: 500
+seed: 3407
+shared_codebook: False
+stride_t: 2
+vq_act: relu
+vq_norm: None
+warm_up_iter: 2000
+weight_decay: 0.0
+which_epoch: all
+width: 512
+window_size: 64
+-------------- End ----------------