Research map / 3D vision / Depth and stereo
Learning Camera Localization via Dense Scene Matching
Shitao Tang, Chengzhou Tang, Rui Huang, Siyu Zhu, Ping Tan
Check 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 |
|---|---|---|---|
| Fuse temporal correlations from neighboring frames at inference (ablation) | DSM · 7scenes and Cambridge Landmarks | pose accuracy: 96.1 (+1.6); 94.5 (+0.7); 99.5 (+3.1); 84.2 (+1.9); 57.2 +(0.2); 69.2 (+0.5); 69.9 (+16.0); 76.7 (+8.2); 84.6 (+1.7); 43 | Improved |
| Sort correlations and select the top K coordinates per pixel (ablation) | DSM · 7scenes | pose accuracy: 0.82 -> 0.96 | Improved |
| Increase the number of scene images from 1 to 5 (ablation) | DSM · 7scenes | pose accuracy: 0.87 -> 0.94 | Improved |
| Use 10 scene images at test time rather than 5, while training with 5 (ablation) | DSM · 7scenes | pose accuracy: 0.94 -> 0.96 | Improved |
| Reduce input resolution from 384×512 to 192×256 (ablation) | DSM · 7scenes | pose accuracy: 0.92 -> 0.96 | Hurt |
| Use temporal correlations in coordinate-map prediction (ablation) | DSM · 7scenes | scene coordinate accuracy at 10 cm threshold | Improved |
| Replace top-K sorted scene candidates with random ball-query sampling | DSM · 7scenes | scene coordinate accuracy | Hurt |
| Use DSM rather than SANet for scene-agnostic localization | scene-agnostic camera localization · 7scenes and Cambridge Landmarks | localization accuracy | Improved |
Closest papers by meaning
- SANet: Scene Agnostic Network for Camera Localization (ICCV 2019 · similarity 0.91)
- Hierarchical Scene Coordinate Classification and Regression for Visual Localization (CVPR 2020 · similarity 0.89)
- Learning Less Is More - 6D Camera Localization via 3D Surface Regression (CVPR 2018 · similarity 0.88)
- Learning to Produce Semi-dense Correspondences for Visual Localization (CVPR 2024 · similarity 0.86)
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.85)
- SLAM3R: Real-Time Dense Scene Reconstruction from Monocular RGB Videos (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.
