Research map / Video / Video segmentation and tracking
CoWTracker: Tracking by Warping instead of Correlation
Zihang Lai, Eldar Insafutdinov, Edgar Sucar, Andrea Vedaldi
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 warping to the tracking head (vs non-warping head) (ablation) | CoWTracker · TAP-Vid DAVIS | w_avg: 54.6 -> 78.0 | Improved |
| add temporal attention to the tracker transformer (ablation) | CoWTracker · RoboTAP | w_avg: 72.6 -> 83.4 | Improved |
| use iterative refinement instead of single pass (ablation) | CoWTracker · TAP-Vid DAVIS | w_avg: 71.4 -> 78.0 | Improved |
| use DPT upsampler instead of none (ablation) | CoWTracker · TAP-Vid DAVIS | w_avg: 72.5 -> 78.0 | Improved |
| use finer indexing stride 1/2 instead of 1/16 (ablation) | CoWTracker · TAP-Vid DAVIS | w_avg: 70.9 -> 78.0 | Improved |
| use VGGT backbone instead of Pi3 (ablation) | CoWTracker · RoboTAP | w_avg: 82.6 -> 83.4 | Improved |
| use VGGT backbone instead of CoTracker ConvNet (ablation) | CoWTracker · TAP-Vid DAVIS | w_avg: 62.9 -> 78.0 | Improved |
| replace AllTracker head with warping head (ablation) | CoWTracker · TAP-Vid DAVIS | w_avg: 72.0 -> 78.0 | Improved |
| use CoWTracker instead of AllTracker (Kub+Mix) | dense point tracking · TAP-Vid/RoboTAP mean | AJ: 68.9 -> 71.3 | Improved |
| use zero-shot CoWTracker vs specialized WAFT (twins-a2) on Spring | optical flow · Spring val | EPE: 0.11 -> 0.17 | Hurt |
| use zero-shot CoWTracker vs WAFT (twins-a2) on KITTI | optical flow · KITTI-2015 train | EPE: 1.15 -> 1.04 | Improved |
Closest papers by meaning
- FlowTrack: Revisiting Optical Flow for Long-Range Dense Tracking (CVPR 2024 · similarity 0.90)
- Dense Optical Tracking: Connecting the Dots (CVPR 2024 · similarity 0.88)
- SpatialTracker: Tracking Any 2D Pixels in 3D Space (CVPR 2024 · similarity 0.88)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.87)
- TrackIME: Enhanced Video Point Tracking via Instance Motion Estimation (NeurIPS 2024 · similarity 0.86)
- TrackCraft3R: Repurposing Video Diffusion Transformers for Dense 3D Tracking (NeurIPS 2026 · similarity 0.86)
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.
