Research map / Detection and segmentation / Tracking
CoTracker: It is Better to Track Together
Nikita Karaev, Ignacio Rocco, Ben Graham, 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 |
|---|---|---|---|
| track points jointly (add cross-track attention and support points) (ablation) | CoTracker · TAP-Vid-DAVIS First | AJ: 55.6 -> 62.2 | Improved |
| track all target points jointly on Dynamic Replica (ablation) | CoTracker · Dynamic Replica | delta_avg^occ: 28.8 -> 37.6 | Improved |
| joint tracking effect on visible points (ablation) | CoTracker · Dynamic Replica | delta_avg^vis: 62.4 -> 68.9 | Improved |
| add unrolled training over sliding windows (ablation) | CoTracker · TAP-Vid-DAVIS First | AJ: 44.6 -> 62.2 | Improved |
| add proxy tokens (64) instead of full self-attention (ablation) | CoTracker · scalability on 80 GB GPU | max tracks / time: 9.4k -> 69.2k tracks; 207.3s -> 27.9s | Improved |
| use 64 proxy tokens instead of 0, 32 or 128 (ablation) | CoTracker · TAP-Vid-DAVIS First | AJ: 0 tokens 61.6; 32: 60.2; 64: 62.2; 128: 60.9 | Mixed |
| add local support point grid (ablation) | CoTracker · TAP-Vid-DAVIS First | AJ: 55.6 -> 60.4 | Improved |
| add global and local support grids together (ablation) | CoTracker · TAP-Vid-DAVIS First | AJ: 55.6 -> 62.2 | Improved |
| use training sliding window 8 instead of 4 or 16 (ablation) | CoTracker · TAP-Vid-DAVIS First | AJ: 4: 56.7; 8: 62.2; 16: 61.1 | Improved |
| evaluate with an inference window different from the trained window size (ablation) | CoTracker · TAP-Vid-DAVIS First | AJ: 62.2 -> 54.9 (window 4) / 54.6 (window 16) | Hurt |
| use CoTracker rather than TAPIR on discontinuous Kinetics videos | CoTracker · TAP-Vid-Kinetics First | AJ: 49.6 -> 48.8 | No clear effect |
Later papers that compared with this method
3 later accepted papers report CoTracker as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Local All-Pair Correspondence for Point Tracking (ECCV 2024 · better on 3 of 4 shared results)
- TrackIME: Enhanced Video Point Tracking via Instance Motion Estimation (NeurIPS 2024 · better on 1 of 1 shared result)
- TAPTR: Tracking Any Point with Transformers as Detection (ECCV 2024 · better on 1 of 1 shared result)
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.
- TAPIR — TAPIR: Tracking Any Point with Per-Frame Initialization and Temporal Refinement (ICCV 2023) · better on 4 of 5 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
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.86)
- TGFormer: Transformer with Track Query Group for Multi-Object Tracking (AAAI 2025 · similarity 0.85)
- RTracker: Recoverable Tracking via PN Tree Structured Memory (CVPR 2024 · similarity 0.85)
- ODTrack: Online Dense Temporal Token Learning for Visual Tracking (AAAI 2024 · similarity 0.85)
- Samba: Synchronized Set-of-Sequences Modeling for Multiple Object Tracking (ICLR 2025 · similarity 0.84)
- SpatialTracker: Tracking Any 2D Pixels in 3D Space (CVPR 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.
