Research map / Video / Video segmentation and tracking
Rethinking Self-Supervised Correspondence Learning: A Video Frame-Level Similarity Perspective
Jiarui Xu, Xiaolong Wang
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 color augmentation (ablation) | VFS ResNet-18 · DAVIS-2017 | J&F_m: 65.0 -> 61.9 | Hurt |
| add color augmentation (ablation) | VFS ResNet-18 · OTB-100 | precision: 48.1 -> 57.3 | Improved |
| sample different frames for predictor and target (distant sampling) (ablation) | VFS ResNet-18 · DAVIS-2017 | J&F_m: 62.2 -> 65.0 | Improved |
| add spatial augmentation (random crop and flip) to different-frame inputs (ablation) | VFS ResNet-18 · OTB-100 | precision: 37.4 -> 48.1 | Improved |
| increase continuous frame interval from 0 to 32 (ablation) | VFS ResNet-18 · OTB-100 | precision: 30.6 -> 46.9 | Improved |
| use distant sampling instead of continuous interval 32 (ablation) | VFS ResNet-18 · DAVIS-2017 | J&F_m: 64.5 -> 65.0 | Improved |
| increase sampled frames from 2 to 8 (no color aug) (ablation) | VFS ResNet-18 · DAVIS-2017 | J&F_m: 65.0 -> 66.7 | Improved |
| increase sampled frames from 2 to 8 (with color aug) (ablation) | VFS ResNet-18 · DAVIS-2017 | J&F_m: 61.9 -> 62.5 | Mixed |
| add negative pairs (contrastive learning) (ablation) | VFS ResNet-18 · OTB-100 | precision: 48.1 -> 39.0 | Hurt |
| add negative pairs (contrastive learning) (ablation) | VFS ResNet-18 · DAVIS-2017 | J&F_m: 65.0 -> 64.7 | Hurt |
| add negative pairs (contrastive learning) (ablation) | VFS ResNet-18 · ImageNet-1k linear probe | top-1 accuracy: 22.0 -> 24.2 | Improved |
| scale backbone from ResNet-18 to ResNet-50 (ablation) | VFS (no negatives, no color aug) · DAVIS-2017 | J&F_m: 65.0 -> 68.9 | Improved |
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.
- SeCo — SeCo: Exploring Sequence Supervision for Unsupervised Representation Learning (AAAI 2021) · better on 1 of 1 shared result
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
- Joint-task Self-supervised Learning for Temporal Correspondence (NeurIPS 2019 · similarity 0.92)
- Modelling Neighbor Relation in Joint Space-Time Graph for Video Correspondence Learning (ICCV 2021 · similarity 0.92)
- Space-Time Correspondence as a Contrastive Random Walk (NeurIPS 2020 · similarity 0.91)
- Contrastive Transformation for Self-supervised Correspondence Learning (AAAI 2021 · similarity 0.91)
- Self-Supervised Cross-View Correspondence with Predictive Cycle Consistency (CVPR 2025 · similarity 0.88)
- Matching Anything by Segmenting Anything (CVPR 2024 · 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.
