Document RoboTwin archive layout
Browse files- RoboTwin/README.md +28 -0
RoboTwin/README.md
ADDED
|
@@ -0,0 +1,28 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# RoboTwin archive layout
|
| 2 |
+
|
| 3 |
+
This directory contains the RoboTwin data used by 4RC.
|
| 4 |
+
|
| 5 |
+
Each task is stored as one uncompressed tar archive:
|
| 6 |
+
|
| 7 |
+
RoboTwin/<task>.tar
|
| 8 |
+
|
| 9 |
+
Every archive contains:
|
| 10 |
+
|
| 11 |
+
<task>/<episode>/images/...
|
| 12 |
+
<task>/<episode>/depths/...
|
| 13 |
+
<task>/<episode>/intrinsics/...
|
| 14 |
+
<task>/<episode>/extrinsics/...
|
| 15 |
+
|
| 16 |
+
Download and reconstruct the original RoboTwin directory:
|
| 17 |
+
|
| 18 |
+
from huggingface_hub import snapshot_download
|
| 19 |
+
|
| 20 |
+
local_dir = snapshot_download(
|
| 21 |
+
repo_id="HarrisonPENG/4RC-Action",
|
| 22 |
+
repo_type="dataset",
|
| 23 |
+
allow_patterns="RoboTwin/*.tar",
|
| 24 |
+
)
|
| 25 |
+
|
| 26 |
+
Then extract all tar files into the same destination directory. The archives
|
| 27 |
+
are not compressed because the source images are already PNG-compressed and
|
| 28 |
+
uncompressed tar supports faster creation and extraction.
|