Research map / Label-efficient and robust learning / Recognition and long-tail
CPlaNet: Enhancing Image Geolocalization by Combinatorial Partitioning of Maps
Paul Hongsuck Seo, Tobias Weyand, Jack Sim, Bohyung Han
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 |
|---|---|---|---|
| combine classifiers via combinatorial partitioning (CPlaNet[1-5] vs naive averaging) | ClassSet 1-5 ensemble · Im2GPS3k | accuracy within 1 km (%): 8.5 -> 9.9 | Improved |
| combinatorial partitioning of two classifiers (CPlaNet[1-2] vs Average[1-2]) | ClassSet 1-2 ensemble · Im2GPS3k | accuracy within 1 km (%): 8.2 -> 9.3 | Improved |
| use five classifiers instead of two in combinatorial partitioning (ablation) | CPlaNet · Im2GPS3k | accuracy within 1 km (%): 9.3 -> 9.9 | Improved |
| add reproduced PlaNet as an additional base classifier (ablation) | CPlaNet[1-5] · Im2GPS3k | accuracy within 1 km (%): 9.9 -> 10.2 | Improved |
| add reproduced PlaNet as an additional base classifier (ablation) | CPlaNet[1-5] · Im2GPS3k | accuracy within 2500 km (%): 64.6 -> 64.6 | No clear effect |
| use visual-distance edge weights instead of geographic-only edge weights for geoclass generation (ablation) | ClassSet · Im2GPS3k | accuracy within 1 km (%): 8.0 -> 8.4 | Improved |
| use PlaNet-style ImageNet features instead of geolocation-trained features (ImageNetFeat vs ClassSet 1) (ablation) | ClassSet 1 · Im2GPS3k | accuracy within 1 km (%): 8.4 -> 3.0 | Hurt |
| CPlaNet[1-5,PlaNet] versus PlaNet (reprod) at coarse scale | CPlaNet · YFCC4k | accuracy within 1 km (%): 5.6 -> 7.9 | Improved |
| CPlaNet[1-5,PlaNet] versus PlaNet (reprod) at 25 km | CPlaNet · YFCC4k | accuracy within 25 km (%): 14.3 -> 14.8 | No clear effect |
| CPlaNet[1-5,PlaNet] versus PlaNet (reprod) at 2500 km | CPlaNet · Im2GPS | accuracy within 2500 km (%): 81.9 -> 78.5 | Hurt |
| CPlaNet[1-5,PlaNet] versus PlaNet (reprod) at street level | CPlaNet · Im2GPS | accuracy within 1 km (%): 11.0 -> 16.5 | Improved |
| add five classification heads instead of one | CPlaNet · Im2GPS3k | inference time per image (s): 0.12 -> 0.13 | No clear effect |
Closest papers by meaning
- PIGEON: Predicting Image Geolocations (CVPR 2024 · similarity 0.88)
- GeoRanker: Distance-Aware Ranking for Worldwide Image Geolocalization (NeurIPS 2025 · similarity 0.88)
- OpenStreetView-5M: The Many Roads to Global Visual Geolocation (CVPR 2024 · similarity 0.88)
- Around the World in 80 Timesteps: A Generative Approach to Global Visual Geolocation (CVPR 2025 · similarity 0.87)
- G3: An Effective and Adaptive Framework for Worldwide Geolocalization Using Large Multi-Modality Models (NeurIPS 2024 · similarity 0.87)
- GeoRouter: Dynamic Paradigm Routing for Worldwide Image Geolocalization (NeurIPS 2026 · similarity 0.87)
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.
