GTA MOT in Sports tackles two of the major problems in player tracking:
- Mix-up errors: one tracklet accidentally containing multiple players
- Cut-off errors: the same player getting a new ID every time they leave and re-enter the camera view
GTA does this in two stages:
- Tracklet Splitter: uses ReID embeddings and DBSCAN to split “mixed” tracklets into identity-pure fragments.
- Tracklet Connector: hierarchically clusters fragments with appearance + temporal + spatial constraints to reconnect long-term identities.
Paper - https://openaccess.thecvf.com/content/ACCV2024W/MLCSA2024/papers/Sun_GTA_Global_Tracklet_Association_for_Multi-Object_Tracking_in_Sports_ACCVW_2024_paper.pdf
Github - https://github.com/sjc042/gta-link
GTA MOT in Sports tackles two of the major problems in player tracking:
GTA does this in two stages:
Paper - https://openaccess.thecvf.com/content/ACCV2024W/MLCSA2024/papers/Sun_GTA_Global_Tracklet_Association_for_Multi-Object_Tracking_in_Sports_ACCVW_2024_paper.pdf
Github - https://github.com/sjc042/gta-link