Research map / 3D vision / Human pose estimation
A Guide to Structureless Visual Localization
Vojtech Panek, Qunjie Zhou, Yaqing Ding, Sérgio Agostinho, Zuzana Kúkelová, Torsten Sattler, Laura Leal-Taixé
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 |
|---|---|---|---|
| use monocular depth (3-point solver) instead of 5-point solver (ablation) | Ess. mat. · Aachen Day-Night v1.1 (day) | localization recall: 59.1 / 80.0 / 97.9 -> 66.7 / 86.8 / 98.4 | Improved |
| use depth-based 3-point solver instead of 5-point solver (ablation) | E5+1 · Aachen Day-Night v1.1 (day) | localization recall: 78.4 / 89.8 / 97.8 -> 54.2 / 63.0 / 70.1 | Hurt |
| use Metric3D versus MASt3R depth predictor (ablation) | Ess. mat. (3Pt+depth) and E3+1 · several datasets | localization recall | No clear effect |
| add semi-generalized pose estimation over multiple database images (E5+1) instead of pairwise relative poses (Ess. mat. 5Pt) | structureless localization · Aachen Day-Night v1.1 (day) | localization recall: 59.1 / 80.0 / 97.9 -> 78.4 / 89.8 / 97.8 | Improved |
| triangulate local 3D points from all retrieved images then P3P instead of semi-generalized E5+1 | E5+1 · Extended CMU Seasons (urban) | localization recall: 93.0 / 97.2 / 99.1 -> 95.5 / 98.5 / 99.5 | Improved |
| triangulate from image pairs instead of from all retrieved images (ablation) | Local triangulation · Extended CMU Seasons (urban) | localization recall: 95.5 / 98.5 / 99.5 -> 18.8 / 68.7 / 99.1 | Hurt |
| triangulate from image pairs instead of all images (ablation) | Local triangulation with MASt3R matches · Extended CMU Seasons | localization recall | Mixed |
| use RoMa dense matcher instead of sparse features (ablation) | structureless methods · outdoor datasets (Aachen, Extended CMU) | localization recall | Improved |
| use MASt3R dense matcher (ablation) | structureless methods · NAVER indoor datasets | localization recall | Improved |
| use RoMa outdoor model instead of indoor model (ablation) | structureless methods · NAVER indoor datasets | localization recall | No clear effect |
| add motion averaging and 2D-3D refinement (LazyLoc) over Ess. mat. (5Pt) | pose triangulation · Aachen Day-Night v1.1 (night) | localization recall: 52.4 / 78.0 / 97.4 -> 57.6 / 83.8 / 95.8 | Mixed |
| increase retrieved images from top-10 to top-20 (ablation) | E5+1 · Aachen Day-Night v1.1 (night) | localization recall: 61.3 / 85.9 / 96.9 -> 65.4 / 85.9 / 96.3 | Improved |
Closest papers by meaning
- Combining Absolute and Semi-Generalized Relative Poses for Visual Localization (IJCV 2026 · similarity 0.94)
- Investigating the Role of Image Retrieval for Visual Localization (IJCV 2022 · similarity 0.90)
- Are Large-Scale 3D Models Really Necessary for Accurate Visual Localization? (CVPR 2017 · similarity 0.90)
- Long-Term Visual Localization Revisited (TPAMI 2020 · similarity 0.89)
- Visual Localization Using Imperfect 3D Models From the Internet (CVPR 2023 · similarity 0.89)
- Understanding the Limitations of CNN-Based Absolute Camera Pose Regression (CVPR 2019 · similarity 0.89)
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.
