Research map / Detection and segmentation / CNN features
From Coarse to Fine: Robust Hierarchical Localization at Large Scale
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 coarse-to-fine retrieval followed by local feature matching instead of direct matching | Hierarchical localization with RootSIFT · Aachen night | recall @ 0.5m/2°: 24.5 -> 30.6 | Helped |
| Use SuperPoint local features instead of RootSIFT | Hierarchical localization with NetVLAD global descriptors · Aachen night | recall @ 0.5m/2°: 30.6 -> 40.8 | Helped |
| Use HF-Net for both global and local features rather than separate NetVLAD and SuperPoint predictors | Hierarchical localization · RobotCar night | recall @ 0.25m/2°: 6.6 -> 2.7 | Hurt |
| Use HF-Net local features rather than SuperPoint with NetVLAD global descriptors (ablation) | Aachen localization · Aachen day | recall @ 0.25m: 79.7 -> 81.2 | Helped |
| Use HF-Net local features rather than SuperPoint with NetVLAD global descriptors (ablation) | Aachen localization · Aachen night | recall @ 0.5m: 40.8 -> 40.8 | No clear effect |
| Use DOAP rather than SuperPoint local features with NetVLAD global descriptors (ablation) | Aachen localization · Aachen day | recall @ 0.5m: 88.0 -> 88.5 | Helped |
| Use DOAP rather than SuperPoint local features with NetVLAD global descriptors (ablation) | Aachen localization · Aachen night | recall @ 0.5m: 40.8 -> 34.7 | Hurt |
| Use HF-Net global descriptors rather than NetVLAD while keeping HF-Net local features (ablation) | Aachen localization · Aachen day | recall @ 0.25m: 81.2 -> 75.7 | Hurt |
| Use HF-Net global descriptors rather than NetVLAD while keeping HF-Net local features (ablation) | Aachen localization · Aachen night | recall @ 5m: 76.5 -> 72.4 | Hurt |
| Train the global retrieval head with night images | HF-Net · Night queries | localization robustness | Helped |
| Use soft keypoint target probabilities instead of hard labels | HF-Net keypoint head · HF-Net training | keypoint performance | Helped |
Closest papers by meaning
- Efficient Large-Scale Localization by Global Instance Recognition (CVPR 2022 · similarity 0.90)
- Hierarchical Scene Coordinate Classification and Regression for Visual Localization (CVPR 2020 · similarity 0.89)
- RELOCATE: A Simple Training-Free Baseline for Visual Query Localization Using Region-Based Representations (CVPR 2025 · similarity 0.87)
- Recognize Any Regions (NeurIPS 2024 · similarity 0.86)
- From Coarse to Fine: A Matching and Alignment Framework for Unsupervised Cross-View Geo-Localization (AAAI 2025 · similarity 0.86)
- R-SCoRe: Revisiting Scene Coordinate Regression for Robust Large-Scale Visual Localization (CVPR 2025 · 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.