AffordPose Dataset Overview

Dataset Name

AffordPose: A Large‑scale Dataset of Hand‑Object Interactions with Affordance‑driven Hand Pose

Dataset Summary

AffordPose is a large‑scale hand‑object interaction dataset focusing on function‑driven hand poses.

Dataset Download

The dataset can be downloaded from the AffordPose project page. Users may download specific categories or the entire dataset as needed. The data is stored as follows:

. └── AffordPose ├── bottle │ ├── 3415 │ │ ├── 3415_Twist │ │ │ ├── 1.json │ │ │ ├── ... │ │ │ └── 28.json │ │ │ │ │ └── 3415_Wrap-grasp │ │ ├── 1.json │ │ ├── ... │ │ └── 28.json | | | └── ... | └── ...

Data File Structure

The structure of data file xxx.json is as follows:

• rhand_mesh # hand mesh
• dofs # hand joint configuration
• rhand_trans # palm translation
• rhand_quat # palm rotation
• object_mesh # object mesh, vertices have functional labels
• trans_obj # default value: (0,0,0)
• quat_obj # default value: (1,0,0,0)
• afford_name # object function corresponding to interaction
• class_name # object category

Data Visualization

• Hand and object meshes can be visualized by saving the rhand_mesh and object_mesh from the xxx.json file as .obj files and opening them in MeshLab.
• The hand model is based on the obman dataset, utilizing the MANO hand model and the GraspIt! simulator.

Citation

If the AffordPose dataset is helpful to your research, please consider citing:

@InProceedings{Jian_2023_ICCV, author = {Jian, Juntao and Liu, Xiuping and Li, Manyi and Hu, Ruizhen and Liu, Jian}, title = {AffordPose: A Large‑Scale Dataset of Hand‑Object Interactions with Affordance‑Driven Hand Pose}, booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)}, month = {October}, year = {2023}, pages = {14713-14724} }