Research map / Video / Video segmentation and tracking
Learning Pixel Trajectories With Multiscale Contrastive Random Walks
Zhangxing Bian, Allan Jabri, Alexei A. Efros, Andrew Owens
Check 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 the regression module to nonparametric multiscale matching (ablation) | Multiscale contrastive random walk · KITTI-15 train | all: 9.42 -> 3.86 | Improved |
| Add edge-aware smoothness loss (ablation) | Full model · KITTI-15 train | all: 3.86 -> 17.43 | Improved |
| Add learned feature consistency to the photometric loss (ablation) | Full model · KITTI-15 train | all: 5.02 -> 3.86 | Improved |
| Keep the regressor constrained near the nonparametric flow estimate (ablation) | Full model · KITTI-15 train | all: 10.44 -> 3.86 | Improved |
| Add the boundary loss (ablation) | Full model · KITTI-15 train | all: 4.88 -> 3.86 | Improved |
| Add the contrastive random walk to the regressor-only setup (ablation) | PWC-Net flow model · KITTI-15 train | all: 21.45 -> 3.86 | Improved |
| Replace Charbonnier photometric loss with learned features (ablation) | PWC-Net flow model · KITTI-15 train | all: 5.69 -> 3.86 | Improved |
| Combine learned features with Census features (ablation) | Photometric loss · KITTI-15 train | all: 3.86 -> 3.10 | Improved |
| Use 3-frame training with learned and Census features (ablation) | Photometric loss model · KITTI-15 train | all: 3.10 -> 3.03 | Improved |
| Increase cycle length from 2 to 4 (ablation) | Multiscale contrastive random walk · KITTI-15 train | all: 3.86 -> 3.39 | Improved |
| Increase the number of pyramid levels from 1 to 5 (ablation) | Multiscale contrastive random walk · KITTI-15 train | all: 8.98 -> 3.86 | Improved |
| Increase local attention window from 3 x 3 to 11 x 11 (ablation) | Multiscale contrastive random walk · KITTI-15 train | all: 7.21 -> 3.68 | Improved |
Closest papers by meaning
- Space-Time Correspondence as a Contrastive Random Walk (NeurIPS 2020 · similarity 0.89)
- Self-Supervised Any-Point Tracking by Contrastive Random Walks (ECCV 2024 · similarity 0.89)
- Contrastive Learning for Space-Time Correspondence via Self-Cycle Consistency (CVPR 2022 · similarity 0.88)
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 2025 · similarity 0.87)
- Emergent Temporal Correspondences from Video Diffusion Transformers (NeurIPS 2025 · similarity 0.86)
- Learning Segmentation from Point Trajectories (NeurIPS 2024 · 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.
