Research map / 3D vision / Depth and stereo
Learning Less Is More - 6D Camera Localization via 3D Surface Regression
Read the paperCheck 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 |
|---|---|---|---|
| replace the scoring CNN with a differentiable soft inlier count (ablation) | DSAC · 7Scenes | pose accuracy below 5cm and 5°: 55.9% to 58.9% | Helped |
| replace the scoring CNN with a differentiable soft inlier count (ablation) | DSAC · 12Scenes | pose accuracy below 5cm and 5°: 79.7% to 89.6% | Helped |
| replace the scoring CNN with a differentiable soft inlier count (ablation) | DSAC · 7Scenes Kitchen | pose accuracy below 5cm and 5°: less than 1% | Helped |
| train using the three-stage procedure, including reprojection-error optimization and end-to-end training (ablation) | Ours · indoor localization | pose accuracy | Helped |
| add reprojection-error optimization after heuristic initialization (ablation) | Ours trained without a 3D model · indoor localization | pose accuracy | Helped |
| add end-to-end training after reprojection-error optimization (ablation) | Ours trained without a 3D model · indoor localization | pose accuracy | Helped |
| use end-to-end training alone, without the preceding initialization and reprojection steps (ablation) | Ours · indoor localization | pose accuracy | Hurt |
| reduce scene-coordinate output resolution from 80x60 to 20x15 (ablation) | Ours · 7Scenes Complete | pose accuracy below 5cm and 5°: 76.1% to 72.8% | Hurt |
| increase scene-coordinate output resolution from 80x60 to 320x240 (ablation) | Ours · 7Scenes Complete | pose accuracy below 5cm and 5°: 76.1% to 76.6% | Helped |
| initialize scene coordinates with a constant depth prior of 10m for 7Scenes (ablation) | Ours trained without a 3D model · 7Scenes | pose accuracy below 5cm and 5°: 49.3% | Hurt |
| train without a 3D scene model (ablation) | Ours · 7Scenes Complete | pose accuracy below 5cm and 5°: 76.1% to 60.4% | Hurt |
| train without a 3D scene model (ablation) | Ours · 12Scenes Complete | pose accuracy below 5cm and 5°: 96.4% to 60.9% | Hurt |
Closest papers by meaning
- Hierarchical Scene Coordinate Classification and Regression for Visual Localization (CVPR 2020 · similarity 0.91)
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.89)
- LOCATE 3D: Real-World Object Localization via Self-Supervised Learning in 3D (ICML 2025 · similarity 0.87)
- HyperPose: Hypernetwork-Infused Camera Pose Localization and an Extended Cambridge Landmarks Dataset (CVPR 2025 · similarity 0.87)
- FAR: Flexible Accurate and Robust 6DoF Relative Camera Pose Estimation (CVPR 2024 · similarity 0.86)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · similarity 0.86)
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.