Research map / Video / Video segmentation and tracking
Learning Correspondence From the Cycle-Consistency of Time
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 skip-cycle loss (ablation) | Cycle-consistency representation · DAVIS-2017 val | J (Mean): 39.5 -> 41.9 | Helped |
| add the skip-cycle loss (ablation) | Cycle-consistency representation · DAVIS-2017 val | F (Mean): 37.9 -> 39.4 | Helped |
| use 5 nearest neighbors instead of 20 for label propagation (ablation) | Ours (ResNet-50) · DAVIS-2017 val | J (Mean): 40.8 -> 41.9 | Helped |
| use 5 nearest neighbors instead of 20 for label propagation (ablation) | Ours (ResNet-50) · DAVIS-2017 val | F (Mean): 38.5 -> 39.4 | Helped |
| use 5 nearest neighbors instead of 10 for label propagation (ablation) | Ours (ResNet-50) · DAVIS-2017 val | J (Mean): 41.5 -> 41.9 | Helped |
| use 5 nearest neighbors instead of 1 for label propagation (ablation) | Ours (ResNet-50) · DAVIS-2017 val | J (Mean): 41.0 -> 41.9 | Helped |
| train on Kinetics instead of VLOG (ablation) | Cycle-consistency representation · DAVIS-2017 val | J (Mean): 41.9 -> 42.5 | No clear effect |
| train on Kinetics instead of VLOG (ablation) | Cycle-consistency representation · DAVIS-2017 val | F (Mean): 39.4 -> 39.2 | No clear effect |
| fine-tune on DAVIS train (ablation) | Ours (ResNet-50) · DAVIS-2017 val | J (Mean): 41.9 -> 42.0 | No clear effect |
| fine-tune on DAVIS train (ablation) | Ours (ResNet-50) · DAVIS-2017 val | F (Mean): 39.4 -> 39.1 | No clear effect |
| learn features by minimizing temporal cycle-consistency on unlabeled video | Ours (ResNet-50) · DAVIS-2017 val | J (Mean): 34.6 -> 41.9 | Helped |
| learn features by minimizing temporal cycle-consistency on unlabeled video | Ours (ResNet-50) · JHMDB split 1 val | PCK@.1: 49.0 -> 57.7 | Helped |
Closest papers by meaning
- Self-Supervised Cross-View Correspondence with Predictive Cycle Consistency (CVPR 2025 · similarity 0.91)
- Joint-task Self-supervised Learning for Temporal Correspondence (NeurIPS 2019 · similarity 0.91)
- Space-Time Correspondence as a Contrastive Random Walk (NeurIPS 2020 · similarity 0.90)
- When the Future Becomes the Past: Taming Temporal Correspondence for Self-supervised Video Representation Learning (CVPR 2025 · 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 0.87)
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.