Dataset Information

The ABC2027 TrackID3x3 Challenge uses the TrackID3x3 dataset, developed for multi-player tracking, player identification, and pose estimation in 3x3 basketball full-court videos.

The dataset URL is here. 

Data update (October 21, 2026): The RKU0 image coordinates in Final challenge/court_registration.csv have been corrected. Please re-download the file.

The dataset contains fixed-camera basketball recordings captured from different viewpoints. Training resources primarily contain top-view videos, while the test and final challenge sets contain previously unseen oblique-view videos recorded at Ryutsu Keizai University (RKU).

The competition uses two sets: the official Training set and the Challenge evaluation set. The Challenge videos have already been shared. There is no separate intermediate Test set or Test-set ranking stage, and each team submits its Challenge predictions once.

The official Training annotations do not overlap with the Challenge ground truth. Publicly available annotations corresponding to the Challenge sequences must not be used for training, pseudo-label generation, or generating predictions.

Policy update (October 23, 2026): Independently generated synthetic data, including those derived from the official Training set or permitted public sources, are allowed, provided that no Challenge-related data or non-public real-world videos/annotations are used.