240 video sequences containing 150,280 frames and 1,646,510 bounding boxes across basketball, football, and volleyball categories. The collection focuses on multi-object tracking in professional sports environments where athletes exhibit rapid, non-linear motion and frequent occlusions.
Use Cases
- Train multi-object tracking models to maintain player identity using the frame index, track ID, and bounding box coordinates
- Evaluate re-identification algorithms on athletes with similar appearances using the sequence-based track ID labels
- Develop motion prediction systems for non-linear movement patterns using the temporal bb_left and bb_top coordinate data
Strengths
- 240 high-definition video sequences totaling 150,280 frames
- 1,646,510 manually annotated bounding boxes for player tracking
- Includes three distinct sports categories: basketball, football, and volleyball
- Features complex player interactions and rapid camera movements typical of broadcast sports