Research map / 3D vision / Depth and stereo
Reference Pose Generation for Long-term Visual Localization via Learned Features and View Synthesis
Zichao Zhang, Torsten Sattler, Davide Scaramuzza
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 |
|---|---|---|---|
| evaluate with refined reference poses instead of original poses (0.25m/2deg is first threshold) | D2-Net · Aachen Day-Night nighttime (98 queries) | queries localized (%): 18.4 -> 74.5 | Improved |
| evaluate with refined reference poses instead of original (5m/10deg) | Active Search v1.1 · Aachen Day-Night nighttime (98 queries) | queries localized (%): 56.1 -> 64.3 | Improved |
| evaluate with refined reference poses instead of original (5m/10deg) | NetVLAD + D2-Net · Aachen Day-Night nighttime (98 queries) | queries localized (%): 85.7 -> 96.9 | Improved |
| evaluate with refined reference poses instead of original (0.25m/2deg) | NetVLAD + D2-Net · Aachen Day-Night nighttime (98 queries) | queries localized (%): 17.3 -> 84.7 | Improved |
| render from current pose and match D2-Net features iteratively instead of D2-Net registration only (ablation) | reference pose generation · Aachen nighttime extension | pose accuracy | Improved |
| use D2-Net features instead of SIFT for real-to-rendered matching (ablation) | reference pose refinement · Aachen nighttime | effective inliers | Improved |
| use D2-Net features instead of R2D2 features (ablation) | reference pose refinement · Aachen nighttime | pose stability | Improved |
| use both MVS mesh and invSfM renderings instead of one (ablation) | reference pose refinement · Aachen nighttime | effective inliers | Improved |
| use MVS mesh rendering instead of invSfM rendering (ablation) | reference pose refinement · Aachen nighttime | effective inliers | Improved |
| run 5 refinement iterations instead of 1 (ablation) | reference pose refinement · Aachen nighttime, perturbed initial poses | success rate | Improved |
| sample only 10% of inlier matches to re-estimate the pose (ablation) | reference pose refinement · Aachen nighttime | pose stability | No clear effect |
Closest papers by meaning
- Long-Term Visual Localization Revisited (TPAMI 2020 · similarity 0.91)
- Benchmarking 6DOF Outdoor Visual Localization in Changing Conditions (CVPR 2018 · similarity 0.88)
- A Guide to Structureless Visual Localization (IJCV 2026 · similarity 0.87)
- R-SCoRe: Revisiting Scene Coordinate Regression for Robust Large-Scale Visual Localization (CVPR 2025 · similarity 0.86)
- Learning to Produce Semi-dense Correspondences for Visual Localization (CVPR 2024 · similarity 0.86)
- 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.
