TUB CrowdFlow is a synthetic video dataset for visual crowd behavior analysis, containing 10 sequences and 5 scenes. Based on drone monitoring, each scene is synthesized twice from a static view and a dynamic view. The resolution of these synthetic videos is 1280×720, the speed is 25 frames/second, and the total number of frames is 3,200, which is also the typical video rendering of current commercial CCTV monitoring systems.

Each sequence has the following ground truth data:

• Optical flow field
• Character tracks (up to 1,451)
• Dense pixel tracks