Research map / Label-efficient and robust learning / Recognition and long-tail
Self-supervising Fine-grained Region Similarities for Large-scale Image Localization
Yixiao Ge, Haibo Wang, Feng Zhu, Rui Zhao, Hongsheng Li
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 top-k difficult positives with hard labels (ablation) | Baseline · Tokyo 24/7 | recall R@1 (%): 80.6 -> 76.2 | Hurt |
| add soft self-enhanced similarity loss to top-k positives training (ablation) | Baseline w/ top-k positives · Tokyo 24/7 | recall R@1 (%): 76.1% to 80.6% | Improved |
| add region augmentation with GPS labels only (ablation) | Baseline · Tokyo 24/7 | recall R@1 (%): 80.6 -> 79.8 | Hurt |
| add all image-to-region similarities (half and quarter regions) (ablation) | Ours · Tokyo 24/7 | recall R@1 (%): 80.6 -> 85.4 | Improved |
| add quarter regions to half regions (ablation) | Ours · Tokyo 24/7 | recall R@1 (%): 84.4 -> 85.4 | Improved |
| add positive sub-regions (ablation) | Ours · Tokyo 24/7 | recall R@1 (%): 80.8 -> 85.4 | Improved |
| add negative sub-regions mining (ablation) | Ours · Tokyo 24/7 | recall R@1 (%): 82.2 -> 85.4 | Improved |
| add k-reciprocal nearest neighbor top-k ranking refinery (ablation) | Ours · Tokyo 24/7 | recall R@1 (%): 83.8 -> 85.4 | Improved |
| add softmax temperature annealing across generations (ablation) | Ours · Tokyo 24/7 | recall R@1 (%): 83.5 -> 85.4 | Improved |
| train more generations (1st to 4th) | Ours · Tokyo 24/7 | recall R@1 (%): 80.6 -> 82.6 -> 84.2 -> 85.4 | Improved |
| replace SARE with proposed method | SARE · Tokyo 24/7 | recall R@1 (%): 79.7 -> 85.4 | Improved |
| replace SARE with proposed method | SARE · Pitts250k-test | recall R@1 (%): 89.0 -> 90.7 | Improved |
Closest papers by meaning
- Divide&Classify: Fine-Grained Classification for City-Wide Visual Geo-Localization (ICCV 2023 · similarity 0.88)
- Scaling Image Geo-Localization to Continent Level (NeurIPS 2025 · similarity 0.88)
- Region Similarity Representation Learning (ICCV 2021 · similarity 0.88)
- MutualVPR: A Mutual Learning Framework for Resolving Supervision Inconsistencies via Adaptive Clustering (NeurIPS 2025 · similarity 0.86)
- From Coarse to Fine: A Matching and Alignment Framework for Unsupervised Cross-View Geo-Localization (AAAI 2025 · similarity 0.86)
- GeoRanker: Distance-Aware Ranking for Worldwide Image Geolocalization (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.
