Publications

End-to-end multi-object trackers have narrowed the gap with classical tracking-by-detection on association-difficult benchmarks. Yet they still make spatially implausible errors no classical tracker would, such as assigning one identity to objects on opposite sides of the frame. A model could learn to avoid them, but tracking annotations are scarce, so we encode spatial priors explicitly instead, while keeping inference fully end-to-end with no post-hoc association.

We propose three spatial priors, at the data, loss, and representation stages. Spatial ID Switches bias trajectory permutations toward spatially overlapping objects, reducing the mismatch between training and inference confusions. Spatial ID Loss scales each identity’s penalty by its box distance, so a distant switch costs more than a nearby one. Spatial Anchor gives each track token its frame position, an explicit spatial cue for attention.

We instantiate the three priors in MOTIP2, a tracker adapted from MOTIP and built on the real-time DEIM detection transformer. Trained without extra data, its main model, MOTIP2-L, sets a new state of the art: 73.4 HOTA on DanceTrack, 76.0 on SportsMOT, and 71.1 IDF1 on PersonPath22. MOTIP2 is a family of models spanning the speed-accuracy trade-off: a lighter model, MOTIP2-S, matches the original MOTIP at over 3× the speed, and MOTIP2-X reaches 74.8 HOTA on DanceTrack.

Multi-object tracking (MOT) requires maintaining consistent identities under occlusions and complex motion, a challenge that end-to-end transformers and segmentation-based trackers address from complementary perspectives. End-to-end approaches benefit from learned identity reasoning but often struggle with association robustness, while segmentation-based methods provide strong object localization but rely on heuristic track management. We propose MOTIC (Multi-Object Tracking as Identity Correction), a simple mechanism that bridges these paradigms by incorporating track-level proposals from a plug-and-play tracker (e.g., SAM3) as structured anchors for identity association within the MOTIP2 framework. By feeding SAM3 proposals into MOTIP2’s identity decoder, MOTIC preserves strong detection quality while substantially improving identity association. On DanceTrack (validation set), MOTIC achieves state-of-the-art performance, surpassing both MOTIP2 and SAM3, improving HOTA by +9.1 and +2.1 points respectively, while also achieving gains in DetA, AssA, MOTA, and IDF1. These results demonstrate the effectiveness of proposal conditioning as a simple yet powerful strategy for improving end-to-end MOT.