Research map / 3D vision / Depth and stereo
Is This the Right Place? Geometric-Semantic Pose Verification for Indoor Visual Localization
Hajime Taira, Ignacio Rocco, Jiri Sedlar, Masatoshi Okutomi, Josef Sivic, Tomas Pajdla, Torsten Sattler, Akihiko Torii
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 |
|---|---|---|---|
| add scan-graph (multiple scans) view synthesis | DensePV · InLoc | pose accuracy (%) at [0.50, 5]: 55.6 -> 59.0 | Improved |
| add scan-graph view synthesis | DensePV · InLoc | pose accuracy (%) at [2.00, 10]: 74.2 -> 71.4 | Hurt |
| add semantic masking (+S) to dense pose verification | DensePV · InLoc (w/o scan-graph) | pose accuracy (%) at [0.50, 5]: 55.6 -> 57.8 | Improved |
| add semantic masking (+S) with scan-graph | DensePV · InLoc (w/ scan-graph) | pose accuracy (%) at [0.25, 5]: 39.8 -> 41.3 | Improved |
| add surface-normal consistency to appearance (DensePNV) | DensePV · InLoc (w/o scan-graph) | pose accuracy (%) at [2.00, 10]: 74.2 -> 76.6 | Improved |
| add surface-normal consistency with scan-graph | DensePV · InLoc (w/ scan-graph) | pose accuracy (%) at [0.50, 5]: 59.0 -> 60.5 | Improved |
| add semantics on top of normals with scan-graph (DensePNV+S) | DensePNV · InLoc (w/ scan-graph) | pose accuracy (%) at [2.00, 10]: 75.4 -> 75.1 | No clear effect |
| use normals only without appearance (DenseNV) | DensePE · InLoc (w/o scan-graph) | pose accuracy (%) at [2.00, 10]: 61.1 -> 62.9 | No clear effect |
| train pose verification network on random perturbed views vs DPE-selected views (ablation) | TrainPV · InLoc | pose accuracy (%) at [2.00, 10]: 76.3 -> 76.3 | No clear effect |
| add trainable verification (TrainPV random) | DensePV · InLoc | pose accuracy (%) at [2.00, 10]: 74.2 -> 76.3 | Improved |
| add image cropping before normal estimation (ablation) | DensePNV · InLoc | pose accuracy (%) at [1.00, 10]: 72.3 -> 71.7 | Hurt |
| add image cropping before normal estimation (ablation) | DenseNV · InLoc | pose accuracy (%) at [0.25, 5]: 32.2 -> 29.5 | Hurt |
Closest papers by meaning
- Pose Correction for Highly Accurate Visual Localization in Large-Scale Indoor Spaces (ICCV 2021 · similarity 0.91)
- InLoc: Indoor Visual Localization With Dense Matching and View Synthesis (CVPR 2018 · similarity 0.88)
- SPVLoc: Semantic Panoramic Viewport Matching for 6D Camera Localization in Unseen Environments (ECCV 2024 · similarity 0.88)
- A Dataset for Benchmarking Image-Based Localization (CVPR 2017 · similarity 0.87)
- Rooms from Motion: Un-posed Indoor 3D Object Detection as Localization and Mapping (NeurIPS 2025 · similarity 0.87)
- Geometric Exploitation for Indoor Panoramic Semantic Segmentation (NeurIPS 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.
