Research map / 3D vision / Depth and stereo
Learning Probabilistic Coordinate Fields for Robust Correspondences
Weiyue Zhao, Hao Lü, Xinyi Ye, Zhiguo Cao, Xin Li
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 confidence map to Cartesian+barycentric coordinates (ablation) | SuperGlue with PCFs (MLP) · PhotoTourism (RootSIFT) | AUC@20: 75.89 -> 76.47 | Improved |
| use attention-based instead of MLP positional encoding (ablation) | SuperGlue with PCFs · PhotoTourism (RootSIFT) | AUC@20: 76.47 -> 77.33 | Improved |
| add Cartesian coordinates to barycentric coordinates (ablation) | SuperGlue with BCF (MLP) · PhotoTourism (RootSIFT) | AUC@20: 74.20 -> 75.89 | Improved |
| replace Cartesian coordinates with PCFs | SuperGlue · PhotoTourism (HardNet) | AUC@5: 32.56 -> 41.22 | Improved |
| replace Cartesian coordinates with PCFs | SuperGlue · SUN3D (RootSIFT) | P_proj: 52.05 -> 51.81 | No clear effect |
| add PCFs to LoFTR | LoFTR DS · MegaDepth | AUC@5: 46.92 -> 48.27 | Improved |
| add PCFs to LoFTR | LoFTR OT · ScanNet | AUC@5: 19.35 -> 20.81 | Improved |
| remove sine positional encoder from LoFTR with PCFs (ablation) | LoFTR DS + PCFs · MegaDepth | AUC@20: 77.53 -> 78.40 | Improved |
| add PCFs to OANet | OANet · YFCC100M (RootSIFT) | precision P_epi: 46.10 -> 57.31 | Improved |
| add PCFs to OANet | OANet · YFCC100M (SuperPoint) | AUC@5: 41.83 -> 41.65 | No clear effect |
| add PCFs to OANet | OANet · SUN3D (RootSIFT) | AUC@5: 17.42 -> 17.33 | No clear effect |
| make sigma+ learnable instead of fixed (ablation) | PCF-Net probabilistic model · Confidence-map IoU (MegaDepth) | IoU: 14.75 -> 23.66 | 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.
- SuperGlue — SuperGlue: Learning Feature Matching With Graph Neural Networks (CVPR 2020) · better on 7 of 7 shared results
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
- A2B: Anchor to Barycentric Coordinate for Robust Correspondence (IJCV 2023 · similarity 0.89)
- Neural Correspondence Field for Object Pose Estimation (ECCV 2022 · similarity 0.86)
- NCP: Neural Correspondence Prior for Effective Unsupervised Shape Matching (NeurIPS 2022 · similarity 0.86)
- COG: Confidence-aware Optimal Geometric Correspondence for Unsupervised Single-reference Novel Object Pose Estimation (CVPR 2026 · similarity 0.86)
- Correspondence learning via linearly-invariant embedding (NeurIPS 2020 · similarity 0.86)
- Learning Affine Correspondences by Integrating Geometric Constraints (CVPR 2025 · 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.
