facebookresearch/ContactPose
A dataset of 2.9M RGB-D grasp images with hand-object contact maps, hand pose, and object pose annotations for computer vision research.

Velocity · 7d
+0.2
★ / day
Trend
→steady
star history
ContactPose is a research dataset and dataloader for hand-object interaction, published at ECCV 2020. It provides synchronized RGB-D images of grasps along with annotated contact maps, 3D hand pose (using the MANO model), and object pose. The repository includes download utilities and Python dataloaders to facilitate training and evaluation of computer vision models for grasp contact prediction and hand pose estimation.