Research map / Video / Video segmentation and tracking
Joint-task Self-supervised Learning for Temporal Correspondence
Read the paperCheck 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 localization during inference (ablation) | Ours on DAVIS-2017 · DAVIS-2017 | J (Mean): 56.3 -> 57.7 | Helped |
| Add the localization module during training (ablation) | Ours on DAVIS-2017 · DAVIS-2017 | J (Mean): 53.8 -> 56.3 | Helped |
| Add orthogonal regularization (ablation) | Ours on DAVIS-2017 · DAVIS-2017 | J (Mean): 55.2 -> 56.3 | Helped |
| Add concentration regularization (ablation) | Ours on DAVIS-2017 · DAVIS-2017 | J (Mean): 48.3 -> 56.3 | Helped |
| Add both orthogonal and concentration regularization (ablation) | Ours on DAVIS-2017 · DAVIS-2017 | J (Mean): 44.3 -> 56.3 | Helped |
| Add localization, orthogonal regularization, and concentration regularization (ablation) | Ours on DAVIS-2017 · DAVIS-2017 | J (Mean): 45.7 -> 56.3 | Helped |
| Jointly learn region localization and fine-grained matching with a shared affinity | Ours versus Wang et al. · DAVIS-2017 | J (Mean): 43.0 -> 56.8 | Helped |
| Use self-supervised correspondence features instead of ImageNet-pretrained ResNet-18 features | Ours versus ResNet-18 (3 blocks) · DAVIS-2017 | J (Mean): 49.4 -> 56.8 | Helped |
| Use joint-task self-supervised features instead of Wang et al. features | Ours versus Wang et al. · VIP | mIoU: 28.9 -> 34.1 | Helped |
| Use joint-task self-supervised features instead of Wang et al. features | Ours versus Wang et al. · J-HMDB | PCK@.1: 57.3 -> 58.6 | Helped |
| Use the proposed features instead of UDT in the tracking framework | Ours versus UDT · OTB2015 | AUC score (%): 59.4 -> 59.2 | No clear effect |
Closest papers by meaning
- Contrastive Transformation for Self-supervised Correspondence Learning (AAAI 2021 · similarity 0.92)
- Self-Supervised Cross-View Correspondence with Predictive Cycle Consistency (CVPR 2025 · similarity 0.89)
- When the Future Becomes the Past: Taming Temporal Correspondence for Self-supervised Video Representation Learning (CVPR 2025 · similarity 0.88)
- No More Shortcuts: Realizing the Potential of Temporal Self-Supervision (AAAI 2024 · similarity 0.88)
- Emergent Temporal Correspondences from Video Diffusion Transformers (NeurIPS 2025 · similarity 0.86)
- Video Diffusion Models Excel at Tracking Similar-Looking Objects Without Supervision (NeurIPS 2025 · 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.