3.6 million 3D human poses and corresponding images captured from 4 high-resolution cameras. The data features 11 professional actors performing 15 distinct daily activities such as 'Greeting', 'Sitting', and 'Walking Dog'.
Use Cases
- Train 3D pose estimation models using the 3D joint coordinates and synchronized camera images
- Evaluate cross-view consistency by comparing 2D projections across the 4 calibrated camera views
- Perform action recognition classification based on the 15 activity labels assigned to each sequence
Strengths
- 3.6 million frames of 3D human poses and synchronized video
- 15 activity classes including 'Phoning', 'Purchasing', and 'Waiting'
- Synchronized video from 4 calibrated cameras captured at 50Hz
- Includes 3D joint positions and 2D projections for each frame