5 video sequences across synthetic and real-world crowd categories with ground truth optical flow and density maps. The data provides a benchmark for motion estimation in high-density environments where occlusion and small-scale motion are prevalent.
Use Cases
- Train optical flow models to handle occlusions using the ground truth flow_x and flow_y vectors
- Benchmark motion estimation accuracy using the provided flow field evaluation metrics
- Develop multi-task models for joint crowd counting and motion estimation using density maps and flow fields
Strengths
- 5 video sequences including synthetic and real-world crowd data
- Ground truth optical flow fields provided for every frame
- Density maps for each frame to support multi-task learning
- Benchmark for evaluating motion estimation in dense pedestrian environments