Research map / Detection and segmentation / Tracking
Collaborative Tracking Learning for Frame-Rate-Insensitive Multi-Object Tracking
Yiheng Liu, Junta Wu, Yi Fu
Read the paperCheck your idea against 137,217 papers
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 collaborative tracking queries without interaction (ablation) | Baseline+E2E · MOT17 val (n=6) | HOTA: 64.5 -> 65.1 | No clear effect |
| add information refinement module (IRM) between decoders (ablation) | BE+CTQ · MOT17 val (n=6) | HOTA: 65.1 -> 66.7 | Improved |
| add IRM (ablation) | BE+CTQ · DanceTrack val (n=6) | HOTA: 59.3 -> 61.0 | Improved |
| use both removal and addition branches in IRM instead of one (ablation) | IRM · MOT17 val (n=6) | HOTA: 66.6 -> 66.7 | Improved |
| add tracking object consistency loss (TOCLoss) (ablation) | BE+CTQ+IRM · DanceTrack val (n=6) | HOTA: 61.0 -> 61.9 | Improved |
| add TOCLoss (ablation) | BE+CTQ+IRM · DanceTrack val (n=6) | MOTA: 87.6 -> 86.5 | Hurt |
| increase max collaborative queries beyond 3 (ablation) | ColTrack · MOT17 val | HOTA | Hurt |
| remove the IRM before the first decoder, keep one before each later decoder (ablation) | ColTrack · MOT17 val | HOTA | Improved |
| use ColTrack instead of Baseline+Bytetrack at low frame rate (n=10) | tracking-by-detection baseline · DanceTrack val (n=10) | HOTA: 26.8 -> 53.7 | Improved |
| use ColTrack at lower frame rate (equivalent FPS) versus YOLOX+Bytetrack | MOT17 val · MOT17 val | equivalent FPS: 27.7 -> 32.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- MOTR — MOTR: End-to-End Multiple-Object Tracking with TRansformer (ECCV 2022) · better on 2 of 2 shared results
- MOTRv2 — MOTRv2: Bootstrapping End-to-End Multi-Object Tracking by Pretrained Object Detectors (CVPR 2023) · better on 1 of 1 shared result
- QDTrack — QDTrack: Quasi-Dense Similarity Learning for Appearance-Only Multiple Object Tracking (TPAMI 2023) · better on 1 of 1 shared result
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
- UniTrack: Differentiable Graph Representation Learning for Multi-Object Tracking (ICLR 2026 · similarity 0.86)
- Focusing on Tracks for Online Multi-Object Tracking (CVPR 2025 · similarity 0.86)
- ODTrack: Online Dense Temporal Token Learning for Visual Tracking (AAAI 2024 · similarity 0.85)
- Sampling-Resilient Multi-Object Tracking (AAAI 2024 · similarity 0.85)
- Less Is More: Token Context-Aware Learning for Object Tracking (AAAI 2025 · similarity 0.85)
- Hybrid-SORT: Weak Cues Matter for Online Multi-Object Tracking (AAAI 2024 · similarity 0.84)
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.
