90,000 multi-view RGB-D frames capturing 3D human-object interactions across 8 subjects and 20 object categories. The dataset includes temporal sequences of SMPL human body meshes and 3D object poses with precise contact annotations.
Use Cases
- Train 3D reconstruction models to predict human and object poses from multi-view RGB images
- Develop contact-aware motion synthesis algorithms using the SMPL mesh and object trajectory data
- Evaluate 3D human-object interaction (HOI) detection using the provided spatial relationship labels
Strengths
- 90,000 frames of synchronized multi-view RGB-D data
- 20 object categories including chairs, tables, and backpacks with corresponding 3D meshes
- Ground truth SMPL human body parameters and 3D object trajectories
- Detailed contact labels between human body vertices and object surfaces