add README
Browse files
README.md
ADDED
|
@@ -0,0 +1,47 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# UCBProject / ProcessedData
|
| 2 |
+
|
| 3 |
+
Phase 1A outputs of the Affordance2Grasp pipeline — used by downstream phases:
|
| 4 |
+
* **Phase 2** (training the main method's PointNet++ contact-prediction net) reads
|
| 5 |
+
`training_fp/{dataset}/{object}.hdf5`.
|
| 6 |
+
* **Phase 3** (inference, grasp sampling, sim execution, Sim2Real deployment) reads
|
| 7 |
+
per-object affordance priors from `human_prior_fp/{object}.hdf5`.
|
| 8 |
+
|
| 9 |
+
The upstream object meshes (SAM3D reconstructions) live in a sibling repo,
|
| 10 |
+
[`UCBProject/ObjMesh`](https://huggingface.co/datasets/UCBProject/ObjMesh).
|
| 11 |
+
|
| 12 |
+
## Layout
|
| 13 |
+
|
| 14 |
+
```
|
| 15 |
+
training_fp/
|
| 16 |
+
oakink/ 100 files (A* C* O* S* Y* — OakInk object IDs)
|
| 17 |
+
dexycb/ 20 files (ycb_dex_01 ... ycb_dex_20 — DexYCB grasp objects)
|
| 18 |
+
human_prior_fp/
|
| 19 |
+
A01001.hdf5 ... Y35037.hdf5 100 oakink files
|
| 20 |
+
ycb_dex_01.hdf5 ... ycb_dex_20.hdf5 20 dexycb files
|
| 21 |
+
```
|
| 22 |
+
|
| 23 |
+
## HDF5 schema
|
| 24 |
+
|
| 25 |
+
### `training_fp/{ds}/{obj}.hdf5` (training-ready)
|
| 26 |
+
| key | shape | dtype | meaning |
|
| 27 |
+
|---|---|---|---|
|
| 28 |
+
| `point_cloud` | (4096, 3) | float32 | 4096 surface samples of the SAM3D object mesh (metric m, mesh canonical frame) |
|
| 29 |
+
| `normals` | (4096, 3) | float32 | per-point unit normals |
|
| 30 |
+
| `human_prior` | (4096,) | float32 | per-point contact probability in [0, 1] (Gaussian-smoothed, per-object max-normalised) |
|
| 31 |
+
| `robot_gt` | (4096,) | float32 | all-zero placeholder (no robot ground truth for this regime) |
|
| 32 |
+
| `force_center` | (3,) | float32 | centroid of mesh verts with `contact_smooth >= 80th percentile` |
|
| 33 |
+
| attr `object` | str | — | object id (e.g. `ycb_dex_14` or `A01001`) |
|
| 34 |
+
|
| 35 |
+
### `human_prior_fp/{obj}.hdf5` (inference-ready, indexed by object only)
|
| 36 |
+
Same five fields as above. Same content as `training_fp/{ds}/{obj}.hdf5`, but flattened
|
| 37 |
+
across datasets so that `inference/predictor.py` can read by object id alone.
|
| 38 |
+
|
| 39 |
+
## Provenance
|
| 40 |
+
|
| 41 |
+
Generated by:
|
| 42 |
+
* `data/batch_depth_pro.py` → `data/batch_haptic.py` → `tools/batch_obj_pose.py` → `data/batch_align_mano_fp.py`
|
| 43 |
+
* Source raw data: DexYCB (NVIDIA, CC-BY-NC-4.0) + OakInk (CVPR 2022).
|
| 44 |
+
* Phase 1A run on lab RTX 5090; OakInk completed 2026-05-10, DexYCB completed 2026-05-14.
|
| 45 |
+
|
| 46 |
+
HO3D-v3 and ARCTIC are not included here — those will be added when the partner runs
|
| 47 |
+
their Phase 1A on those datasets.
|