Research map / Detection and segmentation / Weakly supervised detection and segmentation
Patch2Pix: Epipolar-Guided Pixel-Level Correspondences
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 |
|---|---|---|---|
| Refine NCNet patch proposals into pixel-level matches with regression and confidence filtering | NCNet · HPatches homography estimation, overall, <1 px | accuracy (%): 0.48 -> 0.51 | Helped |
| Use Oracle proposals instead of NCNet proposals | Patch2Pix · HPatches homography estimation, overall, <1 px | accuracy (%): 0.51 -> 0.55 | Helped |
| Use SuperGlue match proposals instead of NCNet proposals | Patch2Pix · InLoc DUC1, localization at 0.25m | localized queries (%): 44.4 -> 50.0 | Helped |
| Use SuperGlue match proposals instead of NCNet proposals | Patch2Pix · InLoc DUC2, localization at 0.25m | localized queries (%): 49.6 -> 57.3 | Helped |
| Use features from layers 0 through 3 rather than only feature 01 (ablation) | Patch2Pix · HPatches homography estimation, overall, <1 px | accuracy (%): 0.37 -> 0.45 | Helped |
| Use features 0123 rather than features 012 (ablation) | Patch2Pix · HPatches homography estimation, overall, <1 px | accuracy (%): 0.34 -> 0.45 | Helped |
| Use mid-level thresholds 50/50 rather than 400/400 (ablation) | Patch2Pix · HPatches homography estimation, overall, <1 px | accuracy (%): 0.43 -> 0.45 | Helped |
| Use fine-level thresholds 5/5 rather than 1/1 (ablation) | Patch2Pix · HPatches homography estimation, overall, <1 px | accuracy (%): 0.43 -> 0.45 | Helped |
| Add patch expansion during training (ablation) | Patch2Pix · HPatches homography estimation, overall, <1 px | accuracy (%): 0.45 -> 0.47 | Helped |
| Use fine-level thresholds 5/5 rather than 15/1 with patch expansion (ablation) | Patch2Pix · HPatches homography estimation, overall, <1 px | accuracy (%): 0.42 -> 0.47 | Helped |
Closest papers by meaning
- PATS: Patch Area Transportation With Subdivision for Local Feature Matching (CVPR 2023 · similarity 0.89)
- PatchMatch-Based Neighborhood Consensus for Semantic Correspondence (CVPR 2021 · similarity 0.86)
- Universal Correspondence Network (NeurIPS 2016 · similarity 0.86)
- CoCosNet v2: Full-Resolution Correspondence Learning for Image Translation (CVPR 2021 · similarity 0.86)
- LoFTR: Detector-Free Local Feature Matching With Transformers (CVPR 2021 · similarity 0.86)
- XFeat: Accelerated Features for Lightweight Image Matching (CVPR 2024 · similarity 0.85)
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.