Research map / Label-efficient and robust learning / Self-supervised and contrastive
Self-Supervised Any-Point Tracking by Contrastive Random Walks
Ayush Shrivastava, Andrew Owens
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 label warping to the contrastive random walk with a transformer (ablation) | GMRW · TAP-Vid Kubric | AJ: 25.4 -> 45.3 | Improved |
| add label warping (ablation) | GMRW · TAP-Vid DAVIS | AJ: 10.4 -> 32.1 | Improved |
| add edge-aware smoothness loss (ablation) | GMRW · TAP-Vid Kubric | AJ: 45.3 -> 49.0 | Improved |
| lower training stride from 4 to 2 (ablation) | GMRW · TAP-Vid DAVIS | AJ: 33.0 -> 34.5 | Improved |
| train on Kinetics instead of Kubric (ablation) | GMRW · TAP-Vid Kubric | AJ: Kubric 47.7 -> 47.5; DAVIS 34.5 -> 34.6 | Mixed |
| decrease evaluation stride from 4 to 1 (ablation) | GMRW · TAP-Vid DAVIS | AJ: 23.5 -> 41.8 | Improved |
| supervise the same architecture with ground-truth tracks instead of self-supervision (ablation) | GMRW architecture · TAP-Vid DAVIS | AJ: supervised 39.1 vs TAP-Net 38.4 | Improved |
| GMRW versus self-supervised baselines across benchmarks | GMRW-C · TAP-Vid Kinetics | AJ: GMRW-C 31.9 vs Flow-Walk-C 40.9 on Kinetics; 41.8 vs 35.2 on DAVIS | Mixed |
Closest papers by meaning
- Learning to LEAP: Efficient Dense Point Tracking by Focusing Where It Matters (AAAI 2026 · similarity 0.90)
- Learning Pixel Trajectories With Multiscale Contrastive Random Walks (CVPR 2022 · similarity 0.89)
- Track-On: Transformer-based Online Point Tracking with Memory (ICLR 2025 · similarity 0.87)
- Self-Supervised Multi-Object Tracking with Path Consistency (CVPR 2024 · similarity 0.86)
- Decoupled Spatio-Temporal Consistency Learning for Self-Supervised Tracking (AAAI 2025 · similarity 0.85)
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 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.
