Research map / 3D vision / Depth and stereo
NormalLoc: Visual Localization on Textureless 3D Models using Surface Normals
Jiro Abe, Gaku Nakano, Kazumine Ogura
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 |
|---|---|---|---|
| train global descriptor and matcher on rendered normals plus real images instead of using pre-trained networks (MeshLoc) | NormalLoc · CADLoc Notre Dame (F) | success ratio, mean DCRE 10%: 23.8 -> 79.9 | Improved |
| use NormalLoc on 3D model with minimal surface geometry | NormalLoc · CADLoc Notre Dame (G) | success ratio, mean DCRE 10%: 4.8 -> 5.8 (10%); 30.7 -> 39.2 (30%) | Mixed |
| use rendered-normal training scheme versus pre-trained AP-GeM and LoFTR on tricolor images | textureless localization on Aachen mesh AC15 · Aachen Day-Night v1.1 night, AC15 | success ratio (0.25m, 2deg): 8.4 -> 47.6 | Improved |
| NormalLoc textureless versus MeshLoc hybrid (real images for global descriptors) on lowest-detail mesh | NormalLoc · Aachen Day-Night v1.1 night, AC13-C | success ratio (0.25m, 2deg): 9.4 -> 36.1 | Improved |
| textureless NormalLoc versus baseline using texture images | NormalLoc · Aachen Day-Night v1.1 night, AC13-C | success ratio (0.25m, 2deg): 63.4 -> 36.1 | Hurt |
| use triplet loss instead of AP-GeM distillation for global descriptor (ablation) | NormalLoc · Aachen Day-Night v1.1, AC13, (5m, 10deg) | success ratio | Improved |
| add AP-GeM distillation global loss (ablation) | NormalLoc · Aachen Day-Night v1.1, AC13, (5m, 10deg) | success ratio | Hurt |
| apply NormalLoc to indoor point clouds | NormalLoc · InLoc DUC1 | success ratio 0.25m @10deg: 47.5 -> 6.1 (vs HLoc+LoFTR) | Hurt |
Closest papers by meaning
- Fast Globally Optimal Surface Normal Estimation from an Affine Correspondence (ICCV 2023 · similarity 0.86)
- InLoc: Indoor Visual Localization With Dense Matching and View Synthesis (CVPR 2018 · similarity 0.86)
- Combining Absolute and Semi-Generalized Relative Poses for Visual Localization (IJCV 2026 · similarity 0.86)
- A Guide to Structureless Visual Localization (IJCV 2026 · similarity 0.86)
- Reloc3r: Large-Scale Training of Relative Camera Pose Regression for Generalizable, Fast, and Accurate Visual Localization (CVPR 2025 · similarity 0.85)
- 360Loc: A Dataset and Benchmark for Omnidirectional Visual Localization with Cross-device Queries (CVPR 2024 · 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.
