Research map / Image generation / Autoregressive and latent generation
AstroLoc: Robust Space to Ground Image Localizer
Gabriele Berton, Alex Stoken, Carlo Masone
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 unsupervised mining loss (Solution 3) to the pair loss (ablation) | AstroLoc · Texas-L | recall@1: 83.6 -> 91.1 | Improved |
| add unsupervised mining loss (Solution 3) to the pair loss (ablation) | AstroLoc · Alps-L | recall@1: 87.2 -> 94.6 | Improved |
| add database clustering loss (Solution 2) to the pair loss (ablation) | AstroLoc · Texas-L | recall@1: 83.6 -> 86.5 | Improved |
| add naive sampling loss (Solution 1) to the pair loss (ablation) | AstroLoc · Texas-L | recall@1: 83.6 -> 84.0 | No clear effect |
| add naive sampling loss (Solution 1) to the pair loss (ablation) | AstroLoc · Alps-L | recall@100: 98.2 -> 98.0 | No clear effect |
| add pair loss to unsupervised mining alone (ablation) | AstroLoc · Texas-L | recall@1: 82.2 -> 91.1 | Improved |
| sample clusters by query distribution (Solution 3) instead of uniform clustering (Solution 2), loss alone (ablation) | AstroLoc · Texas-L | recall@1: 72.4 -> 82.2 | Improved |
| use AstroLoc training instead of EarthLoc++ (same architecture) | EarthLoc++ · Texas (EarthLoc test set) | recall@1: 80.0 -> 96.1 | Improved |
| use AstroLoc training instead of EarthLoc++ (same architecture) | EarthLoc++ · Texas-L (extended test set) | recall@1: 72.4 -> 91.1 | Improved |
| use AstroLoc-tiny (27M parameters, 512-d) instead of full AstroLoc | AstroLoc · Lost-in-Space (VINSat queries) | recall@1: 52.7 -> 36.7 | Hurt |
| use AstroLoc on worldwide search instead of EarthLoc | EarthLoc · Worldwide search (881k database) | recall@100: 66.6 -> 96.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- EarthLoc++ — EarthLoc: Astronaut Photography Localization by Indexing Earth from Space (CVPR 2024) · better on 17 of 17 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
- EarthLoc: Astronaut Photography Localization by Indexing Earth from Space (CVPR 2024 · similarity 0.94)
- GeoRanker: Distance-Aware Ranking for Worldwide Image Geolocalization (NeurIPS 2025 · similarity 0.85)
- COSMIC: Compress Satellite Image Efficiently via Diffusion Compensation (NeurIPS 2024 · similarity 0.84)
- PIGEON: Predicting Image Geolocations (CVPR 2024 · similarity 0.84)
- LightLoc: Learning Outdoor LiDAR Localization at Light Speed (CVPR 2025 · similarity 0.83)
- Scaling Image Geo-Localization to Continent Level (NeurIPS 2025 · similarity 0.83)
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.
