Add base GraspXL archives

README.md

@@ -1,3 +1,239 @@
----
-license: cc-by-nc-4.0
----
+---
+license: cc-by-nc-4.0
+pretty_name: GraspXL
+size_categories:
+- 10M<n<100M
+tags:
+- grasping
+- dexterous-hand
+- motion-generation
+- objaverse
+- physics-simulation
+- robotics
+- 3d
+---
+
+# GraspXL
+
+GraspXL is a large-scale dataset of physically plausible dexterous grasping
+motion sequences over Objaverse objects. It contains generated grasping motions
+for different hand models, together with processed object meshes used by the
+recorded sequences.
+
+## Why use GraspXL?
+
+### Summary on the dataset
+
+- It contains 10M+ diverse grasping motions for 500k+ objects of different
+  dexterous hands.
+- All grasping motions are generated with physics simulation, which helps ensure
+  the physical plausibility of the generated motions.
+- Each motion contains accurate object and hand poses for every frame.
+
+### Potential tasks with GraspXL
+
+- Generating general grasping motions for
+  [full-body motion generation](https://eth-ait.github.io/phys-fullbody-grasp/).
+- Achieving zero-shot text-to-motion generation with off-the-shelf
+  [text-to-mesh](https://dreamfusion3d.github.io/) generation methods.
+- Generating large-scale pseudo 3D RGBD grasping motions with
+  [texture generation](https://mq-zhang1.github.io/HOIDiffusion/) methods, which
+  can support downstream applications such as
+  [training a general pose estimation model](https://nvlabs.github.io/FoundationPose/).
+- [Simulating general human hand motions](https://eth-ait.github.io/synthetic-handovers/)
+  for human-robot interaction.
+- Serving as expert demonstrations for robot imitation learning.
+
+## Dataset instruction
+
+The dataset is composed of several `.zip` archives. These archives contain the
+generated diverse grasping motion sequences for different hands on
+[Objaverse](https://objaverse.allenai.org/), as well as the processed object mesh
+files. The object meshes are scaled and decimated before use.
+
+To make the dataset easier to download, recorded motion sequences are split into
+several `.zip` files. For most objects, each sequence archive contains 5
+sequences, so users can choose which archives to download.
+
+## Files
+
+The repository contains the following archive groups:
+
+| File | Description |
+| --- | --- |
+| `object_dataset.zip` | Processed object meshes used by the recorded sequences. |
+| `allegro_dataset_1.zip` | Allegro Hand grasping motion sequences. |
+| `allegro_dataset_2.zip` | Additional Allegro Hand grasping motion sequences. |
+| `mano_dataset_1.zip.part00`, `mano_dataset_1.zip.part01` | Split parts of `mano_dataset_1.zip`, containing MANO hand grasping motion sequences. |
+| `mano_dataset_2.zip.part00`, `mano_dataset_2.zip.part01` | Split parts of `mano_dataset_2.zip`, containing additional MANO hand grasping motion sequences. |
+| `mano_dataset_3.zip.part00`, `mano_dataset_3.zip.part01` | Split parts of `mano_dataset_3.zip`, containing additional MANO hand grasping motion sequences. |
+| `leap_dataset_1.zip` | LEAP hand grasping motions for a subset of the objects. |
+| `objaverse_urdf.zip` | URDF files of the Objaverse objects used by the simulator. |
+
+## Objects
+
+`object_dataset.zip` is organized as follows:
+
+```text
+object_dataset.zip
+|-- small
+|   |-- <object_id>
+|   |   `-- <object_id>.obj
+|   `-- ...
+|-- medium
+|   |-- <object_id>
+|   |   `-- <object_id>.obj
+|   `-- ...
+`-- large
+    |-- <object_id>
+    |   `-- <object_id>.obj
+    `-- ...
+```
+
+`small`, `medium`, and `large` contain object meshes with different scales used
+by the recorded sequences. Please check the paper for more details.
+
+## Allegro sequences
+
+`allegro_dataset_1.zip` is organized as follows:
+
+```text
+allegro_dataset_1.zip
+|-- small
+|   |-- <object_id>
+|   |   |-- allegro_1.npy
+|   |   |-- allegro_2.npy
+|   |   |-- allegro_3.npy
+|   |   `-- ...
+|   `-- ...
+|-- medium
+|   |-- <object_id>
+|   |   |-- allegro_1.npy
+|   |   `-- ...
+|   `-- ...
+`-- large
+    |-- <object_id>
+    |   |-- allegro_1.npy
+    |   `-- ...
+    `-- ...
+```
+
+Not every object has the same number of recorded sequences.
+
+Each `.npy` file contains a single motion sequence:
+
+```python
+data = np.load("allegro_x.npy", allow_pickle=True).item()
+```
+
+The sequence dictionary contains:
+
+- `data["right_hand"]["trans"]`: NumPy array of shape `(frame_num, 3)`. Wrist
+  position sequence.
+- `data["right_hand"]["rot"]`: NumPy array of shape `(frame_num, 3)`. Wrist
+  orientation sequence in axis-angle representation.
+- `data["right_hand"]["pose"]`: NumPy array of shape `(frame_num, 22)`. The
+  first 6 dimensions of each frame are 0, and the remaining 16 dimensions are
+  joint angles.
+- `data[object_id]["trans"]`: NumPy array of shape `(frame_num, 3)`. Object
+  position sequence.
+- `data[object_id]["rot"]`: NumPy array of shape `(frame_num, 3)`. Object
+  orientation sequence in axis-angle representation.
+- `data[object_id]["angle"]`: Reserved field, not used.
+
+`allegro_dataset_2.zip` follows the same format and contains another group of
+recorded motion sequences.
+
+## MANO sequences
+
+The MANO archives are stored as split files to keep each repository file under
+the hosting file-size limit. Reconstruct the original archives before extracting:
+
+```bash
+cat mano_dataset_1.zip.part00 mano_dataset_1.zip.part01 > mano_dataset_1.zip
+cat mano_dataset_2.zip.part00 mano_dataset_2.zip.part01 > mano_dataset_2.zip
+cat mano_dataset_3.zip.part00 mano_dataset_3.zip.part01 > mano_dataset_3.zip
+```
+
+`mano_dataset_1.zip` is organized as follows:
+
+```text
+mano_dataset_1.zip
+|-- small
+|   |-- <object_id>
+|   |   |-- mano_1.npy
+|   |   |-- mano_2.npy
+|   |   |-- mano_3.npy
+|   |   `-- ...
+|   `-- ...
+|-- medium
+|   |-- <object_id>
+|   |   |-- mano_1.npy
+|   |   `-- ...
+|   `-- ...
+`-- large
+    |-- <object_id>
+    |   |-- mano_1.npy
+    |   `-- ...
+    `-- ...
+```
+
+Not every object has the same number of recorded sequences.
+
+Each `.npy` file contains a single motion sequence:
+
+```python
+data = np.load("mano_x.npy", allow_pickle=True).item()
+```
+
+The sequence dictionary contains:
+
+- `data["right_hand"]["trans"]`: NumPy array of shape `(frame_num, 3)`. Wrist
+  position sequence.
+- `data["right_hand"]["rot"]`: NumPy array of shape `(frame_num, 3)`. Wrist
+  orientation sequence in axis-angle representation. For MANO, these are the
+  first 3 dimensions of the MANO parameter.
+- `data["right_hand"]["pose"]`: NumPy array of shape `(frame_num, 45)`. The
+  remaining 45 dimensions of the MANO parameter.
+- `data[object_id]["trans"]`: NumPy array of shape `(frame_num, 3)`. Object
+  position sequence.
+- `data[object_id]["rot"]`: NumPy array of shape `(frame_num, 3)`. Object
+  orientation sequence in axis-angle representation.
+- `data[object_id]["angle"]`: Reserved field, not used.
+
+`mano_dataset_2.zip` and `mano_dataset_3.zip` follow the same format and contain
+additional groups of recorded motion sequences.
+
+## LEAP sequences
+
+`leap_dataset_1.zip` contains LEAP hand grasping motions for a subset of the
+500k+ objects. The dataset structure and data format are the same as the Allegro
+Hand data described above.
+
+## Additional data
+
+`objaverse_urdf.zip` contains the URDF files of the 500k+ Objaverse objects.
+These URDF files are used by the code to generate the motions in simulation.
+
+## Loading a sequence
+
+Example:
+
+```python
+import numpy as np
+
+data = np.load("allegro_1.npy", allow_pickle=True).item()
+object_id = next(key for key in data.keys() if key != "right_hand")
+
+right_hand_trans = data["right_hand"]["trans"]
+right_hand_rot = data["right_hand"]["rot"]
+right_hand_pose = data["right_hand"]["pose"]
+
+object_trans = data[object_id]["trans"]
+object_rot = data[object_id]["rot"]
+```
+
+## License
+
+This dataset is released under the
+[Creative Commons Attribution-NonCommercial 4.0 International License](https://creativecommons.org/licenses/by-nc/4.0/).

allegro_dataset_1.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0ad39d1a9c74571d10302b5342dac3eaff9de27e96475379424c11c8df3f359b
+size 36881265777

allegro_dataset_2.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6df1ef4c6f5165e0b060bf7c4a3d3485fed558d2311ebe8133405b8ee1dae1cc
+size 36264468335

leap_dataset_1.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab21eeafdd5e6e4ae4788507ffda1258288631484cbfb4a9639f721937715959
+size 3056233115

objaverse_urdf.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e6f099de75cff856978e71c706cbd6ee6d0345a99d0af12759f42de51413549
+size 49436690151

object_dataset.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4f2948df64455d6e8eb4ebb71cd3940287e802f9aca937114a2683261d0e0bf
+size 19776001955