Research map / Detection and segmentation / CNN features
LoFTR: Detector-Free Local Feature Matching With Transformers
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 |
|---|---|---|---|
| use detector-free Transformer matching with coarse-to-fine refinement | local feature matching · HPatches | Homography estimation AUC @3px: 53.9 -> 65.9 | Helped |
| use detector-free Transformer matching with coarse-to-fine refinement | local feature matching · ScanNet | Pose estimation AUC @5°: 16.16 -> 22.06 | Helped |
| use detector-free Transformer matching with coarse-to-fine refinement | local feature matching · MegaDepth | Pose estimation AUC @5°: 42.18 -> 52.8 | Helped |
| use LoFTR-OT rather than SP+SuperGlue | HLoc visual localization · Aachen v1.1 full localization, day | Localization success (0.5m,5°): 96.1 -> 95.6 | Hurt |
| use LoFTR-OT rather than SP+SuperGlue | HLoc visual localization · InLoc DUC1 | Localization success (0.5m,10°): 68.7 -> 72.2 | Helped |
| use LoFTR-OT rather than SP+SuperGlue | HLoc visual localization · InLoc DUC2 | Localization success (0.5m,10°): 77.1 -> 74.8 | Hurt |
| add global self- and cross-attention instead of convolution (ablation) | LoFTR on ScanNet · ScanNet | Pose estimation AUC @5°: 14.98 -> 20.06 | Helped |
| use the finer 1/8 coarse and 1/2 fine feature resolutions (ablation) | LoFTR on ScanNet · ScanNet | Pose estimation AUC @5°: 16.75 -> 20.06 | Helped |
| add positional encoding once rather than at every Transformer layer (ablation) | LoFTR on ScanNet · ScanNet | Pose estimation AUC @5°: 18.02 -> 20.06 | Helped |
| increase the layer counts to Nc=8 and Nf=2 (ablation) | LoFTR on ScanNet · ScanNet | Pose estimation AUC @10°: 40.8 -> 40.23 | No clear effect |
Closest papers by meaning
- Efficient LoFTR: Semi-Dense Local Feature Matching with Sparse-Like Speed (CVPR 2024 · similarity 0.92)
- ETO:Efficient Transformer-based Local Feature Matching by Organizing Multiple Homography Hypotheses (NeurIPS 2024 · similarity 0.88)
- XFeat: Accelerated Features for Lightweight Image Matching (CVPR 2024 · similarity 0.86)
- JamMa: Ultra-lightweight Local Feature Matching with Joint Mamba (CVPR 2025 · similarity 0.86)
- Unifying Feature and Cost Aggregation with Transformers for Semantic and Visual Correspondence (ICLR 2024 · similarity 0.85)
- Focus on Local: Finding Reliable Discriminative Regions for Visual Place Recognition (AAAI 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.