Research map / Detection and segmentation / Tracking
PathTrack: Fast Trajectory Annotation With Path Supervision
Santiago Manen, Michael Gygli, Dengxin Dai, Luc Van Gool
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 |
|---|---|---|---|
| train the LP tracker's person matcher on PathTrack instead of MOT15 | LP tracker · PathTrack test set | MOTA: 24.5 -> 27.6 | Improved |
| train the LP tracker's person matcher on PathTrack instead of MOT15 | LP tracker · PathTrack test set | ID Switches: 1,827 -> 1,576 | Improved |
| add PathTrack-trained person matcher (TRID) to relink NOMTwSDP trajectories | NOMTwSDP · MOT15 test | ID Switches: 427 -> 351 | Improved |
| add PathTrack-trained person matcher (TRID) to relink NOMTwSDP trajectories | NOMTwSDP · MOT15 test | MOTA: 55.5 -> 55.7 | Improved |
| add PathTrack-trained person matcher (TRID) to relink NOMTwSDP trajectories | NOMTwSDP · MOT15 test | FP: 5,594 -> 6,273 | Hurt |
| train person matching on PathTrack (15,380 trajectories) instead of MOT15 (521) | Siamese CNN · PathTrack test set | matching accuracy: 78% -> 88% | Improved |
| train with PathTrack when context features are used | Siamese CNN + context features · PathTrack test set | matching accuracy: 84% -> 90% | Improved |
| use high-quantity strategy (140 PathTrack videos) instead of high-quality (11 MOT15 videos) at equal annotation time (ablation) | Siamese CNN · PathTrack test set | matching accuracy: 78% -> 85% | Improved |
| increase number of training trajectories toward 15,380 (ablation) | Siamese CNN · PathTrack test set | matching accuracy | Improved |
| add path supervision to box annotations | LabelMe/VATIC box annotation · MOT15 train annotation accuracy | recall at IoU | Improved |
| add path supervision versus box-only annotation, large budgets | VATIC / LabelMe · MOT15 train annotation accuracy | recall at IoU | No clear effect |
| use path supervision instead of box annotation for the same accuracy | VATIC / LabelMe · MOT15 train | annotation time: almost twice and three times more time | Improved |
Closest papers by meaning
- MotionTrack: Learning Robust Short-Term and Long-Term Motions for Multi-Object Tracking (CVPR 2023 · similarity 0.92)
- TAO: A Large-Scale Benchmark for Tracking Any Object (ECCV 2020 · similarity 0.90)
- DanceTrack: Multi-Object Tracking in Uniform Appearance and Diverse Motion (CVPR 2022 · similarity 0.90)
- MOTS: Multi-Object Tracking and Segmentation (CVPR 2019 · similarity 0.90)
- Multiple Object Tracking as ID Prediction (CVPR 2025 · similarity 0.89)
- Self-Supervised Multi-Object Tracking with Path Consistency (CVPR 2024 · similarity 0.88)
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.
