Research map / Image generation / Autoregressive and latent generation
GeoCLIP: Clip-Inspired Alignment between Locations and Images for Effective Worldwide Geo-localization
Vicente Vivanco Cepeda, Gaurav Kumar Nayak, Mubarak Shah
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 |
|---|---|---|---|
| Use Equal Earth Projection rather than direct coordinate input (ablation) | GeoCLIP location encoder · Im2GPS3k | accuracy within 25 km: 11.01 -> 17.08 | Improved |
| Add random Fourier features to EEP and MLP GPS encoding (ablation) | GeoCLIP location encoder · Im2GPS3k | accuracy within 1 km: 0.23 -> 5.37 | Improved |
| Add a dynamic GPS negative queue to EEP and RFF (ablation) | GeoCLIP location encoder · Im2GPS3k | accuracy within 1 km: 5.37 -> 9.84 | Improved |
| Combine coarse, medium, and fine location representations (ablation) | GeoCLIP · Im2GPS3k | accuracy within 1 km: 13.61 -> 14.11 | Improved |
| Reduce training data to 20% | GeoCLIP · Im2GPS3k | accuracy within 2500 km: 83.8 -> 83.8 | No clear effect |
| Replace static queue with dynamic queue (ablation) | GeoCLIP contrastive training · Im2GPS3k | accuracy within 1 km: 5.93 -> 7.55 | Improved |
| Add 1000 m Gaussian noise to dynamic queue coordinates (ablation) | GeoCLIP contrastive training · Im2GPS3k | accuracy within 750 km: 67.46 -> 69.07 | Improved |
| Increase dynamic queue noise beyond 1000 m (ablation) | GeoCLIP contrastive training · Im2GPS3k | accuracy within 2500 km: 83.35 -> 82.87 | Hurt |
| Increase GPS gallery from 21K to 100K coordinates (ablation) | GeoCLIP retrieval · Im2GPS3k | accuracy within 1 km: 11.88 -> 14.11 | Improved |
| Increase GPS gallery from 100K to 500K coordinates (ablation) | GeoCLIP retrieval · YFCC26k | accuracy within 1 km: 11.61 -> 11.49 | Hurt |
| Use a 500K rather than 21K GPS gallery (ablation) | GeoCLIP retrieval · GWS15k | accuracy within 1 km: 0.18 -> 0.6 | Improved |
| Set queue length to 4096 rather than 512 (ablation) | GeoCLIP contrastive training · Im2GPS3k | accuracy within 1 km: 7.57 -> 9.84 | Improved |
Closest papers by meaning
- GeoRanker: Distance-Aware Ranking for Worldwide Image Geolocalization (NeurIPS 2025 · similarity 0.92)
- G3: An Effective and Adaptive Framework for Worldwide Geolocalization Using Large Multi-Modality Models (NeurIPS 2024 · similarity 0.90)
- GeoRouter: Dynamic Paradigm Routing for Worldwide Image Geolocalization (NeurIPS 2026 · similarity 0.89)
- SatCLIP: Global, General-Purpose Location Embeddings with Satellite Imagery (AAAI 2025 · similarity 0.89)
- Vision-Language Reasoning for Geolocalization: A Reinforcement Learning Approach (AAAI 2026 · similarity 0.89)
- PIGEON: Predicting Image Geolocations (CVPR 2024 · similarity 0.88)
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.
