Research map / Detection and segmentation / Tracking
Uncertainty-aware Unsupervised Multi-Object Tracking
Kai Liu, Sheng Jin, Zhihang Fu, Ze Chen, Rongxin Jiang, Jieping Ye
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add uncertainty-aware tracklet labeling (UTL) to long-term-association baseline (ablation) | U2MOT baseline · MOT17 val (half split) | HOTA: 63.43 -> 63.84 | Improved |
| add vanilla long-term similarity association for pseudo identities (LTD) (ablation) | U2MOT baseline · MOT17 val (half split) | HOTA: 63.40 -> 63.43 | No clear effect |
| add tracklet-guided augmentation (TGA) on top of UTL (ablation) | U2MOT · MOT17 val (half split) | HOTA: 63.84 -> 64.08 | Improved |
| use uncertainty-based source and target anchor selection in TGA instead of random (ablation) | TGA · MOT17 val (half split) | HOTA: 63.93 -> 64.08 | Improved |
| choose high-uncertainty target anchor instead of random (ablation) | TGA · MOT17 val (half split) | HOTA: 63.93 -> 64.01 | Improved |
| add TGA with fully random anchor selection (ablation) | U2MOT with UTL · MOT17 val (half split) | HOTA: 63.84 -> 63.93 | Improved |
| use unsupervised U2MOT instead of identity-supervised training | U2MOT · MOT17 val (half split) | HOTA: 63.96 -> 64.08 | Improved |
| apply UTL association at inference to existing trackers | FairMOT · MOT17 val (half split) | HOTA: 62.03 -> 64.01 | Improved |
| apply UTL association at inference to existing trackers | DeepSORT · MOT17 val (half split) | HOTA: 58.48 -> 59.75 | Improved |
| apply UTL association at inference | ByteTrack · MOT17 val (half split) | IDS: 207 -> 212 | Hurt |
| replace pre-trained UniTrack ReID with jointly trained ReID head | ByteTrack on VisDrone-MOT · VisDrone-MOT test-dev | IDF1: 68.3 -> 69.0 | Improved |
| use U2MOT on crowded MOT20 versus ByteTrack | U2MOT · MOT20 test | IDS: 1223 -> 1379 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ByteTrack — ByteTrack: Multi-Object Tracking by Associating Every Detection Box (ECCV 2022) · better on 6 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Probabilistic Tracklet Scoring and Inpainting for Multiple Object Tracking (CVPR 2021 · similarity 0.86)
- Self-Supervised Multi-Object Tracking with Path Consistency (CVPR 2024 · similarity 0.84)
- Focusing on Tracks for Online Multi-Object Tracking (CVPR 2025 · similarity 0.84)
- Samba: Synchronized Set-of-Sequences Modeling for Multiple Object Tracking (ICLR 2025 · similarity 0.83)
- S2-Track: A Simple yet Strong Approach for End-to-End 3D Multi-Object Tracking (ICML 2025 · similarity 0.83)
- TGFormer: Transformer with Track Query Group for Multi-Object Tracking (AAAI 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
