ARC-SORT reports 81.82 HOTA on SportsMOT test tracking
Original title: ARC-SORT: Adaptive, Robust, Compensated Multi-Object Tracking with Interacting Multiple Models
The tracker trains nothing, runs about 50 frames per second on a laptop RTX 3060, and its code is not yet released.
ARC-SORT is a tracking-by-detection method for sports broadcasts in which athletes change direction while the camera pans and zooms. An interacting multiple-model Kalman filter sets each track's search region, and close crossings can receive one corrective identity swap. The authors report 81.82 HOTA and tied-third MOTA of 97.09 on SportsMOT, with the fewest identity switches among 69 entries as of August 2026, plus 62.1 HOTA on DanceTrack.
Unconfirmed: Leaderboard rank and scores are the authors' figures as of August 2026.
- The identity-swap review adds +1.0 HOTA on SportsMOT and DanceTrack.
- On Deep-EIoU detections it reaches 80.31 HOTA, +3.1 over Deep-EIoU.
- Reappearing players are relinked offline by appearance, without a motion gate.
- Corrective swaps add under a millisecond per frame.
From the original article
1 min read at huggingface.co
Multi-object tracking in sports broadcasts is hard for two reasons: athletes move fast and change direction abruptly, and the camera pans and zooms at the same time. We present ARC-SORT (Adaptive, Robust, Compensated SORT), a tracking-by-detection tracker that trains nothing and runs in real time. Its design rule: the tracker measures its own uncertainty, and that measurement decides how it acts.
By Girish Kumar Adari · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.