Update dataset card for tarball layout
Browse files
README.md
ADDED
|
@@ -0,0 +1,69 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: other
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
tags:
|
| 6 |
+
- robotics
|
| 7 |
+
- grasping
|
| 8 |
+
- manipulation
|
| 9 |
+
- 3d-meshes
|
| 10 |
+
- usd
|
| 11 |
+
- simulation
|
| 12 |
+
- simvla
|
| 13 |
+
- bodex
|
| 14 |
+
size_categories:
|
| 15 |
+
- 100K<n<1M
|
| 16 |
+
pretty_name: SimVLA-BODex
|
| 17 |
+
---
|
| 18 |
+
|
| 19 |
+
# SimVLA-BODex
|
| 20 |
+
|
| 21 |
+
Preprocessed object meshes packaged for use with the
|
| 22 |
+
[SimVLA](https://github.com/exaFLOPs26/SimVLA) project. The contents are a
|
| 23 |
+
re-distribution of meshes generated by the
|
| 24 |
+
[BODex](https://github.com/JYChen18/BODex) preprocessing pipeline.
|
| 25 |
+
|
| 26 |
+
## Contents
|
| 27 |
+
|
| 28 |
+
The dataset ships as **3 tarballs** (~2.1 GB compressed, ~6.3 GB extracted)
|
| 29 |
+
to keep the per-file count low and avoid HuggingFace tree-listing rate
|
| 30 |
+
limits on download.
|
| 31 |
+
|
| 32 |
+
| Tarball | Compressed | Contents |
|
| 33 |
+
|---|---|---|
|
| 34 |
+
| `graspdata_final.tar.gz` | ~1.6 GB | Grasp pose data |
|
| 35 |
+
| `processed_data.tar.gz` | ~0.5 GB | Full object set, 2,339 dirs |
|
| 36 |
+
| `use_data.tar.gz` | ~28 MB | Curated 129-object kitchen subset — what `kitchen_scene_generator.py` actually uses |
|
| 37 |
+
|
| 38 |
+
After extraction, each `core_*/` object dir contains:
|
| 39 |
+
- `mesh/simplified.obj` — collision-friendly simplified mesh
|
| 40 |
+
- `mesh/normalized.obj` — normalized mesh
|
| 41 |
+
- `mesh/material_0.png`, `mesh/material.mtl` — material/texture
|
| 42 |
+
- `urdf/coacd_decomposed.urdf` + supporting `.usd` / `.usda` — physics-ready convex decomposition
|
| 43 |
+
- `info/` — per-object metadata
|
| 44 |
+
|
| 45 |
+
## Use with SimVLA
|
| 46 |
+
|
| 47 |
+
The SimVLA repo provides a download helper that fetches and extracts the
|
| 48 |
+
tarballs transparently:
|
| 49 |
+
|
| 50 |
+
```bash
|
| 51 |
+
# Full bundle (all 3 tarballs):
|
| 52 |
+
export BODEX_OBJ_DIR=/path/to/your/bodex
|
| 53 |
+
python scripts/tools/download_bodex.py
|
| 54 |
+
|
| 55 |
+
# Or just the curated kitchen subset:
|
| 56 |
+
python scripts/tools/download_bodex.py --allow-patterns 'use_data.tar.gz'
|
| 57 |
+
```
|
| 58 |
+
|
| 59 |
+
After it finishes you'll have `$BODEX_OBJ_DIR/{graspdata_final,processed_data,use_data}/`.
|
| 60 |
+
See [SimVLA's installation guide](https://github.com/exaFLOPs26/SimVLA/blob/main/docs/installation.md#7-scene-synthesizer)
|
| 61 |
+
for the scene-synthesizer config that points at `${BODEX_OBJ_DIR}/use_data`.
|
| 62 |
+
|
| 63 |
+
## License
|
| 64 |
+
|
| 65 |
+
The underlying object meshes and grasp poses derive from BODex's processing
|
| 66 |
+
of upstream sources. See the [BODex repository](https://github.com/JYChen18/BODex)
|
| 67 |
+
for the license terms applicable to the underlying meshes. This redistribution
|
| 68 |
+
is provided to streamline reproducibility for SimVLA users; if you need to use
|
| 69 |
+
the meshes outside SimVLA, please consult the upstream license.
|