Research map / Video / Video segmentation and tracking
Mining Better Samples for Contrastive Learning of Temporal Correspondence
Sangryul Jeon, Dongbo Min, Seungryong Kim, Kwanghoon Sohn
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 |
|---|---|---|---|
| Use the proposed confidence-based positive mining and dynamic semi-hard negative mining | pixel-level video correspondence learning · DAVIS-2017 validation | J&Fmean: 67.6 -> 70.3 | Improved |
| Use the full proposed method | pixel-level video correspondence learning · YouTube-VOS validation | Overall: 64.2 -> 67.3 | Improved |
| Add temporal coherence to the positive-match confidence pipeline (ablation) | the proposed method · DAVIS-2017 validation | J&Fmean: 68.7 -> 70.3 | Improved |
| Add optimal transport refinement to matching confidence (ablation) | the proposed method · DAVIS-2017 validation | J&Fmean: 65.1 -> 70.3 | Improved |
| Add forward-backward consistency to matching confidence (ablation) | the proposed method · DAVIS-2017 validation | J&Fmean: 63.7 -> 70.3 | Improved |
| Add negative samples to positive-only training (ablation) | the proposed method · DAVIS-2017 validation | J&Fmean: 64.6 -> 70.3 | Improved |
| Select semi-hard negatives rather than all possible negatives (ablation) | the proposed method · DAVIS-2017 validation | J&Fmean: 66.4 -> 70.3 | Improved |
| Adapt the negative-sample hardness threshold with a dynamic curriculum (ablation) | the proposed method · DAVIS-2017 validation | J&Fmean: 67.6 -> 70.3 | Improved |
| Use a dynamic rather than linear negative-mining curriculum (ablation) | the proposed method · DAVIS-2017 validation | J&Fmean: 68.9 -> 70.3 | Improved |
Closest papers by meaning
- Conditional Negative Sampling for Contrastive Learning of Visual Representations (ICLR 2021 · similarity 0.87)
- Learning Contrastive Representation for Semantic Correspondence (IJCV 2022 · similarity 0.87)
- Contrastive Transformation for Self-supervised Correspondence Learning (AAAI 2021 · similarity 0.87)
- Rethinking Self-Supervised Correspondence Learning: A Video Frame-Level Similarity Perspective (ICCV 2021 · similarity 0.86)
- $\mathbb{X}$-Sample Contrastive Loss: Improving Contrastive Learning with Sample Similarity Graphs (ICLR 2025 · similarity 0.86)
- Contrastive Learning with Hard Negative Samples (ICLR 2021 · 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.
