Research map / Label-efficient and robust learning / Recognition and long-tail
Universal Correspondence Network
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 hard negative mining (active, on-the-fly) to UCN training (ablation) | UCN · PASCAL-Berkeley | mean PCK@0.1: +2.6 (36.0 to 38.6) | Helped |
| add convolutional spatial transformer to UCN with hard negative mining (ablation) | UCN · PASCAL-Berkeley | mean PCK@0.1: +5.4 (38.6 to 44.0) | Helped |
| add convolutional spatial transformer to UCN with hard negative mining (ablation) | UCN · FlowWeb PASCAL 12 rigid classes | mean PCK@0.05: +6.3 (36.7 to 43.0) | Helped |
| add convolutional spatial transformer to UCN with hard negative mining (ablation) | UCN · MPI-Sintel and KITTI Flow 2015 | PCK@10px: -0.8 Sintel, -3.1 KITTI | Hurt |
| add convolutional spatial transformer to UCN (ablation) | UCN · KITTI raw camera motion | translation deviation (deg): 4.147 to 4.728 (worse) | Hurt |
| use UCN (raw nearest-neighbor, no global optimization) instead of prior methods with global optimization | UCN · MPI-Sintel and KITTI Flow 2015 | PCK@10px: Sintel 91.5 vs best prior 89.2; KITTI 86.5 vs 85.6 | Helped |
| train UCN with correspondence contrastive loss and hard negative mining, compared to conv4 flow on PASCAL | UCN · PASCAL-Berkeley | mean PCK@0.1: 44.0 vs 24.9 (conv4 flow) | Helped |
| add hard negative mining and spatial transformer to UCN for camera motion via essential matrix | UCN · KITTI raw camera motion | translation deviation (deg): Ours-HN 4.147 vs SIFT 4.749 | Helped |
| add hard negative mining to random-negative training (ablation) | UCN · PASCAL-Berkeley | mean PCK@0.025: +1.6 (11.5 to 13.1) | Helped |
Closest papers by meaning
- SFNet: Learning Object-Aware Semantic Correspondence (CVPR 2019 · similarity 0.90)
- Deep Semantic Feature Matching (CVPR 2017 · similarity 0.90)
- Convolutional Hough Matching Networks (CVPR 2021 · similarity 0.89)
- SemAlign3D: Semantic Correspondence between RGB-Images through Aligning 3D Object-Class Representations (CVPR 2025 · similarity 0.88)
- Jamais Vu: Exposing the Generalization Gap in Supervised Semantic Correspondence (NeurIPS 2025 · similarity 0.87)
- Pixel-level Semantic Correspondence through Layout-aware Representation Learning and Multi-scale Matching Integration (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.