Research map / Graph learning / Graph algorithms and clustering
GlueStick: Robust Image Matching by Sticking Points and Lines Together
Rémi Pautrat, Iago Suárez, Yifan Yu, Marc Pollefeys, Viktor Larsson
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 |
|---|---|---|---|
| match points and lines jointly in one GNN instead of lines only (ablation) | GlueStick · ScanNet homography | median pose error (deg): 27.6 -> 14.1 | Improved |
| replace SuperGlue-based endpoint matching with GlueStick joint wireframe network | points+lines matching · ScanNet homography | median pose error (deg): 17.1 -> 14.1 | Improved |
| add lines to GlueStick keypoints on 7Scenes Stairs (ablation) | GlueStick · 7Scenes Stairs | pose accuracy at 5cm/5deg (%): 55.4 -> 79.7 | Improved |
| add lines on InLoc (ablation) | GlueStick · InLoc | pose AUC | Improved |
| add wireframe preprocessing and line message passing to SuperGlue backbone (ablation) | SuperGlue line matching · ETH3D line matching | average precision | Improved |
| remove keypoints (line-only GlueStick-L) (ablation) | GlueStick · ETH3D line matching | average precision | Hurt |
| use attention instead of average in line message passing (ablation) | LMP · design choice | performance | No clear effect |
| fine-tune on MegaDepth after homography pre-training (ablation) | GlueStick · ETH3D line matching | precision-recall | Improved |
| reduce number of keypoints to zero (ablation) | GlueStick line matching · homography validation set | recall and AP: decrease by at most 5% | Hurt |
| add lines for HPatches homography (ablation) | GlueStick · HPatches | homography AUC: 77.41 -> 78.14 (5px) | No clear effect |
| add lines to points for rotation estimation (ablation) | GlueStick · SUN360 | AUC at 0.5 deg: 0.860 -> 0.890 | Improved |
| estimate pose from essential matrix instead of homography | GlueStick · ScanNet | median pose error (deg): 14.1 -> 8.4 | Improved |
Closest papers by meaning
- Combinatorial Learning of Robust Deep Graph Matching: An Embedding Based Approach (TPAMI 2020 · similarity 0.86)
- FC-GNN: Recovering Reliable and Accurate Correspondences from Interferences (CVPR 2024 · similarity 0.86)
- Structured Set Matching Networks for One-Shot Part Labeling (CVPR 2018 · similarity 0.86)
- DGC-GNN: Leveraging Geometry and Color Cues for Visual Descriptor-Free 2D-3D Matching (CVPR 2024 · similarity 0.86)
- Deep Matching Prior: Test-Time Optimization for Dense Correspondence (ICCV 2021 · similarity 0.86)
- MATCHA: Towards Matching Anything (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.
