Research map / Label-efficient and robust learning / Recognition and long-tail
EDM: Efficient Deep Feature Matching
Xi Li, Tong Rao, Cihui Pan
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 Correlation Injection Module transformer layers (L=2 vs L=0) (ablation) | EDM · MegaDepth ablation (1152x1152) | AUC@5deg: 51.8 -> 57.5 | Improved |
| use L=2 instead of L=4 (ablation) | EDM · MegaDepth ablation (1152x1152) | AUC@5deg / time: 57.5 -> 57.6; 86.0 ms -> 101.9 ms | No clear effect |
| use deep-narrow feature extractor instead of shallow-wide (ablation) | EDM · MegaDepth ablation (1152x1152) | AUC@5deg / time: 56.9 -> 57.5; 109.8 ms -> 86.0 ms | Improved |
| use QKNA instead of vanilla attention (ablation) | EDM · MegaDepth ablation (1152x1152) | AUC@5deg: 56.7 -> 57.5 | Improved |
| use Injection Layers instead of element-wise sum (ablation) | EDM · MegaDepth ablation (1152x1152) | AUC@5deg: 55.9 -> 57.5 | Improved |
| use top-K coarse selection instead of MNN (ablation) | EDM · MegaDepth ablation (1152x1152) | AUC@5deg / time: 57.2 -> 57.5; 101.2 ms -> 86.0 ms | Improved |
| use focal loss instead of negative log-likelihood for coarse matching (ablation) | EDM · MegaDepth ablation (1152x1152) | AUC@5deg: 56.3 -> 57.5 | Improved |
| add bidirectional refinement (ablation) | EDM · MegaDepth ablation (1152x1152) | AUC@5deg: 55.7 -> 57.5 | Improved |
| add fine selection by sigma (ablation) | EDM · MegaDepth ablation (1152x1152) | AUC@5deg: 57.0 -> 57.5 | Improved |
| use Laplace instead of Gaussian prior distribution (ablation) | EDM · MegaDepth ablation (1152x1152) | AUC@5deg: 57.1 -> 57.5 | Improved |
| add Soft Coordinate Classification (ablation) | EDM · MegaDepth ablation (1152x1152) | AUC@5deg: 56.5 -> 57.5 | Improved |
| use RLE loss instead of L1 loss (ablation) | EDM · MegaDepth ablation (1152x1152) | AUC@5deg: 53.9 -> 57.5 | 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.
- TopicFM — TopicFM: Robust and Interpretable Topic-Assisted Feature Matching (AAAI 2023) · better on 1 of 1 shared result
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
- Image Matching from Handcrafted to Deep Features: A Survey (IJCV 2020 · similarity 0.88)
- ETO:Efficient Transformer-based Local Feature Matching by Organizing Multiple Homography Hypotheses (NeurIPS 2024 · similarity 0.87)
- Deep Matching Prior: Test-Time Optimization for Dense Correspondence (ICCV 2021 · similarity 0.87)
- JamMa: Ultra-lightweight Local Feature Matching with Joint Mamba (CVPR 2025 · similarity 0.85)
- EDM: Equirectangular Projection-Oriented Dense Kernelized Feature Matching (CVPR 2025 · similarity 0.85)
- Unifying Feature and Cost Aggregation with Transformers for Semantic and Visual Correspondence (ICLR 2024 · similarity 0.84)
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.
