Research map / 3D vision / Depth and stereo
LightGlue: Local Feature Matching at Light Speed
Philipp Lindenberger, Paul-Edouard Sarlin, Marc Pollefeys
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 matchability classifier (disentangle similarity and matchability) (ablation) | LightGlue · Synthetic homographies | precision (%): 67.4 -> 86.8 | Improved |
| use rotary relative positional encoding instead of absolute (ablation) | LightGlue · Synthetic homographies | precision (%): 84.2 -> 86.8 | Improved |
| use bidirectional cross-attention instead of full cross-attention (ablation) | LightGlue · Synthetic homographies | run time (ms): 22.8 -> 19.4 ms, precision 86.6 vs 86.8 | Improved |
| add deep supervision so earlier layers predict matches (ablation) | LightGlue · Synthetic homographies | recall (%): >90% recall after 5 layers | Improved |
| add adaptive depth and width (ablation) | LightGlue · MegaDepth-1500 (SuperPoint, RANSAC) | time (ms): 44.2 -> 31.4 ms, AUC@5 49.9 -> 49.4 | Improved |
| adaptive depth and width on easy vs hard pairs (ablation) | LightGlue · MegaDepth pairs by overlap | speedup over non-adaptive: 1.86 (easy) vs 1.16 (hard) | Mixed |
| replace SuperGlue with LightGlue | SuperPoint matching · Aachen Day-Night | pairs per second: 6.5 -> 17.2 | Improved |
| replace SuperGlue with LightGlue | SuperPoint matching · Synthetic homographies | precision (%): 74.6 -> 86.8 | Improved |
| replace SuperGlue with LightGlue | SuperPoint matching · MegaDepth-1500 RANSAC AUC@5 | pose AUC @5deg: 49.7 -> 49.9 | No clear effect |
| use DISK features with LightGlue instead of SuperPoint | LightGlue · IMC 2020 stereo AUC@5 | AUC @5deg: 59.03 -> 67.02 | Improved |
| use LightGlue instead of dense ASpanFormer | feature matching · MegaDepth-1500 RANSAC AUC@5 | pose AUC @5deg: 55.3 -> 49.9 | Hurt |
Later papers that compared with this method
6 later accepted papers report LightGlue as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Collaborative Feature Matching with Progressive Correspondence Learning (AAAI 2026 · better on 4 of 4 shared results)
- Learning Dense Feature Matching via Lifting Single 2D Image to 3D Space (ICCV 2025 · better on 2 of 2 shared results)
- Correspondences of the Third Kind: Camera Pose Estimation from Object Reflection (ECCV 2024 · better on 4 of 4 shared results)
- Fully Geometric Panoramic Localization (CVPR 2024 · better on 0 of 3 shared results)
- SRPose: Two-view Relative Pose Estimation with Sparse Keypoints (ECCV 2024 · better on 1 of 2 shared results)
- Raising the Ceiling: Conflict-Free Local Feature Matching with Dynamic View Switching (ECCV 2024 · better on 1 of 1 shared result)
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 8 of 8 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
- SuperGlue: Learning Feature Matching With Graph Neural Networks (CVPR 2020 · similarity 0.89)
- OmniGlue: Generalizable Feature Matching with Foundation Model Guidance (CVPR 2024 · similarity 0.87)
- ETO:Efficient Transformer-based Local Feature Matching by Organizing Multiple Homography Hypotheses (NeurIPS 2024 · similarity 0.86)
- JamMa: Ultra-lightweight Local Feature Matching with Joint Mamba (CVPR 2025 · similarity 0.86)
- Efficient LoFTR: Semi-Dense Local Feature Matching with Sparse-Like Speed (CVPR 2024 · similarity 0.86)
- Matching While Perceiving: Enhance Image Feature Matching with Applicable Semantic Amalgamation (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.
