Research map / Label-efficient and robust learning / Self-supervised and contrastive
Do Different Tracking Tasks Require Different Appearance Models?
Zhongdao Wang, Hengshuang Zhao, Ya-Li Li, Shengjin Wang, Philip Torr, Luca Bertinetto
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 |
|---|---|---|---|
| Use UniTrack ImageNet appearance features instead of FairMOT's task-trained appearance model | MOT-16 tracking · MOT-16 test | IDs: 1074 to 683 | Improved |
| Use a shared UniTrack appearance model instead of LightTrack | PoseTrack · PoseTrack2018 val | IDF1: +21.0 points | Improved |
| Use UniTrack instead of LightTrack | PoseTrack · PoseTrack2018 val | MOTA: 1.3 points | Hurt |
| Use UniTrack ImageNet DCF head instead of XCorr head (ablation) | UniTrack box propagation · OTB-2015 | AUC: 55.5 -> 61.8 | Improved |
| Use UniTrack DCF head instead of SiamFC | Single-object tracking · OTB-2015 | AUC: 3.6 AUC points | Improved |
| Use UniTrack ImageNet mask propagation instead of SiamMask | Video object segmentation · DAVIS-2017 | J-mean: 4.1 J-mean points | Improved |
| Use RSM instead of global-pooled features (ablation) | UniTrack association · MOT-16 train, DPM detections, motion cues | IDF1: 34.4 -> 39.6 | Improved |
| Use RSM instead of supervised ReID features (ablation) | UniTrack association · MOT-16 train, FRCNN detections, appearance only | IDF1: 47.0 -> 51.5 | Improved |
| Use a shared UniTrack representation instead of COSTA | MOTS tracking · MOTS test | IDF1: 70.3 vs 68.2 | Hurt |
| Use a shared UniTrack representation instead of COSTA | MOTS tracking · MOTS test | IDs: 421 vs 342 | Improved |
| Use a shared UniTrack representation instead of KeyTrack | PoseTrack · PoseTrack2018 val | MOTA: 66.6 vs 63.3 | Hurt |
| Use UniTrack VFS instead of ImageNet features | Video object segmentation · DAVIS-2017 | J-mean: 58.4 -> 62.8 | Improved |
Closest papers by meaning
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 2025 · similarity 0.85)
- VastTrack: Vast Category Visual Object Tracking (NeurIPS 2024 · similarity 0.85)
- Foundation Model Driven Appearance Extraction for Robust Multiple Object Tracking (AAAI 2025 · similarity 0.85)
- Beyond Accuracy: Tracking more like Human via Visual Search (NeurIPS 2024 · similarity 0.84)
- SPMTrack: Spatio-Temporal Parameter-Efficient Fine-Tuning with Mixture of Experts for Scalable Visual Tracking (CVPR 2025 · similarity 0.84)
- SUTrack: Towards Simple and Unified Single Object Tracking (AAAI 2025 · 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.
