CrowdTrack-MOT

CrowdTrack-MOT Pedestrian multi-object-tracking benchmark in MOTChallenge format, derived from Loseevaya/CrowdTrack. 33 CCTV / surveillance sequences, single class person, ~702k GT boxes, 1920×1080. What changed vs. the original Converted to MOT format and added a seqinfo.ini per sequence. The original ships two heterogeneous annotation formats (a few sequences as plain gt.txt, the rest as per-frame LabelMe JSON) — both are unified into a single MOT gt.txt.… See the full description on the dataset page:

Source: Hugging Face Hub (Fleyderer/CrowdTrack-MOT). Metadata imported from the dataset’s Hub tags.