Research map / 3D vision / Human pose estimation
Pose Correction for Highly Accurate Visual Localization in Large-Scale Indoor Spaces
Janghun Hyeon, Joohyung Kim, Nakju Doh
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 pose correction (re-estimate pose with reorganized local features) (ablation) | coarse-to-fine localization baseline · InLoc DUC2 | accuracy within 1.0 m: 87.0 -> 93.9 | Improved |
| add pose correction (ablation) | coarse-to-fine localization baseline · InLoc DUC1 | accuracy within 1.0 m: 85.9 -> 89.4 | Improved |
| add pose correction (ablation) | coarse-to-fine localization baseline · M-site | accuracy within 0.5 m: 65.9 -> 68.9 | Improved |
| add pose update step to candidates (ablation) | baseline with 10 candidates · InLoc DUC2 | accuracy within 0.5 m: 82.4 -> 90.1 | Improved |
| add reranking by inlier count (K2 = 20) (ablation) | pose correction · InLoc DUC2 | accuracy within 0.25 m: 67.2 -> 67.9 | Improved |
| use modified pose verification (MPV) instead of PV (ablation) | proposed pipeline · InLoc DUC2 | accuracy within 1.0 m: 90.8 -> 93.9 | Improved |
| use modified pose verification (MPV) instead of PV (ablation) | baseline pipeline · InLoc DUC1 and DUC2 | accuracy within 1.0 m: DUC1 85.9 -> 88.4, DUC2 87.0 -> 85.5 | Mixed |
| add divided matching instead of full matching (ablation) | pose correction without inter-pose · InLoc DUC1 and DUC2 | accuracy within 0.5 m / 0.25 m: DUC2 0.25 m 67.9 -> 69.5, DUC1 0.5 m 76.8 -> 75.8 | Mixed |
| add inter-pose matching (ablation) | pose correction with divided matching and normal filtering · InLoc DUC2 | accuracy within 0.25 m: 67.2 -> 71.0 | Improved |
| add VLF map (ablation) | pose correction with full matching · InLoc DUC1 | accuracy within 1.0 m: 89.4 -> 90.4 | Improved |
| add point normal filtering (ablation) | pose correction with VLF map and divided matching · InLoc DUC1 and DUC2 | accuracy within 1.0 m: DUC2 92.4 -> 93.9, DUC1 90.9 -> 89.9 | Mixed |
| increase SuperPoint features from 3,000 to 4,096 | pose correction · InLoc DUC1 and DUC2 | accuracy within 1.0 m: DUC1 89.4 -> 90.4, DUC2 93.9 -> 93.1 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- InLoc [54] — InLoc: Indoor Visual Localization With Dense Matching and View Synthesis (CVPR 2018) · better on 3 of 3 shared results
- SuperGlue — SuperGlue: Learning Feature Matching With Graph Neural Networks (CVPR 2020) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- InLoc: Indoor Visual Localization With Dense Matching and View Synthesis (CVPR 2018 · similarity 0.91)
- Is This the Right Place? Geometric-Semantic Pose Verification for Indoor Visual Localization (ICCV 2019 · similarity 0.91)
- Large-Scale Localization Datasets in Crowded Indoor Spaces (CVPR 2021 · similarity 0.90)
- RELOCATE: A Simple Training-Free Baseline for Visual Query Localization Using Region-Based Representations (CVPR 2025 · similarity 0.87)
- 360Loc: A Dataset and Benchmark for Omnidirectional Visual Localization with Cross-device Queries (CVPR 2024 · similarity 0.86)
- Rooms from Motion: Un-posed Indoor 3D Object Detection as Localization and Mapping (NeurIPS 2025 · 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.
