1,000+ video sequences across 10 sports categories featuring 3D human motion capture and fine-grained semantic action labels. The dataset provides per-frame 3D pose, shape parameters, and semantic attributes for athletes in challenging motion scenarios.
Use Cases
- Train 3D human mesh recovery models using the SMPL pose and shape parameters
- Develop fine-grained action recognition systems using the semantic action labels
- Benchmark monocular 3D pose estimation algorithms against ground-truth motion capture data in sports contexts
Strengths
- Includes 10 distinct sports categories such as gymnastics, tennis, and basketball
- Provides SMPL model parameters for 3D human body pose and shape estimation
- Contains fine-grained semantic action labels for detailed behavioral understanding
- Features high-dynamic monocular video sequences from real-world sports broadcasts