Research map / Label-efficient and robust learning / Semi-supervised learning
CoTracker3: Simpler and Better Point Tracking by Pseudo-Labelling Real Videos
Nikita Karaev, Yuri Makarov, Jianyuan Wang, Natalia Neverova, Andrea Vedaldi, Christian Rupprecht
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 |
|---|---|---|---|
| fine-tune on 15k real pseudo-labelled videos | CoTracker3 online · TAP-Vid mean | delta_avg^vis: 75.1 -> 76.1 | Improved |
| add cross-track attention (ablation) | CoTracker3 online · Dynamic Replica | delta_avg^occ: 35.9 -> 41.0 | Improved |
| add cross-track attention (ablation) | CoTracker3 online · Dynamic Replica | delta_avg^vis: 71.3 -> 72.9 | Improved |
| self-training on own predictions using real videos (ablation) | CoTracker3 online · TAP-Vid mean | AJ: 62.2 -> 63.5 | Improved |
| add more diverse teachers (RT online, RT offline, TAPIR, CoTracker) (ablation) | pseudo-labelling · TAP-Vid mean | delta_avg: 75.7 -> 76.8 | Improved |
| choose a random teacher instead of averaging or median of teachers (ablation) | pseudo-labelling · TAP-Vid DAVIS | delta_avg: 76.5 -> 77.0 | Improved |
| freeze confidence and visibility head during real-video fine-tuning (ablation) | CoTracker3 · TAP-Vid average | OA: 86.3 -> 90.2 | Improved |
| change point sampling method (Uniform, SuperPoint, DISK, SIFT) (ablation) | pseudo-label queries · TAP-Vid | delta_avg: DAVIS 76.7 to 77.0 | No clear effect |
| scale real video data beyond 30k videos | CoTracker3 online and offline, LocoTrack · TAP-Vid | delta_avg | No clear effect |
| repeat the scaling pipeline a second time (ablation) | CoTracker3 offline · TAP-Vid average | delta_avg: 76.8 -> 76.9 | No clear effect |
| train longer on 15k videos (60k vs 15k iterations) (ablation) | CoTracker3 offline · TAP-Vid average | delta_avg: 76.8 -> 77.0 | No clear effect |
| add real-video fine-tuning to offline model for occluded points | CoTracker3 online · Dynamic Replica | delta_avg^occ: online 41.0 -> 40.1, offline 41.8 -> 42.3 | Mixed |
Later papers that compared with this method
1 later accepted paper reports CoTracker3 as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- MegaFlow: Zero-Shot Large Displacement Optical Flow (ECCV 2026 · 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
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.86)
- TrackCraft3R: Repurposing Video Diffusion Transformers for Dense 3D Tracking (NeurIPS 2026 · similarity 0.86)
- DriveTrack: A Benchmark for Long-Range Point Tracking in Real-World Videos (CVPR 2024 · similarity 0.85)
- TrackIME: Enhanced Video Point Tracking via Instance Motion Estimation (NeurIPS 2024 · similarity 0.85)
- Is This Tracker On? A Benchmark Protocol for Dynamic Tracking (NeurIPS 2025 · similarity 0.85)
- Tracktention: Leveraging Point Tracking to Attend Videos Faster and Better (CVPR 2025 · similarity 0.85)
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.
