Research map / Vision-language and multimodal / Human and embodied modeling
Where We Are and What We're Looking At: Query Based Worldwide Image Geo-Localization Using Hierarchies and Scenes
Brandon Clark, Alec Kerrigan, Parth Parag Kulkarni, Vicente Vivanco Cepeda, 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 |
|---|---|---|---|
| Hierarchy-dependent scene-query decoder (scene queries with max-confidence selection) vs. no scene prediction (ablation) | Swin-based GeoDecoder · Im2GPS3k | accuracy @ 1 km: +1.1 (11.7 to 12.8) | Improved |
| Scene-query selection vs. scene prediction as a secondary task (ablation) | Swin-based GeoDecoder · YFCC26k | accuracy @ 1 km: +0.4 (9.7 to 10.1) | Improved |
| Replace Swin encoder with ViT encoder in the full model (ablation) | GeoDecoder · YFCC26k | accuracy @ 1 km: -1.4 (10.1 to 8.7) | Hurt |
| Use 2 hierarchy-dependent decoder layers vs. 0 (fully hierarchy-independent) (ablation) | GeoDecoder · Im2GPS3k | accuracy @ 1 km: +0.6 (12.2 to 12.8) | Improved |
| Use 2 hierarchy-dependent decoder layers vs. 0 (ablation) | GeoDecoder · YFCC26k | accuracy @ 1 km: +0.4 (9.7 to 10.1) | Improved |
| Increase hierarchy-dependent layers from 2 to 6 (ablation) | GeoDecoder · YFCC26k | accuracy @ 1 km: -1.4 (10.1 to 8.7) | Hurt |
| GeoDecoder depth 3 to 8 layers (ablation) | GeoDecoder · Im2GPS3k | accuracy @ 1 km: +0.9 (11.9 to 12.8) | Improved |
| GeoDecoder depth 8 to 10 layers (ablation) | GeoDecoder · Im2GPS3k | accuracy @ 1 km: -0.3 (12.8 to 12.5) | Hurt |
| Train with 7 geographic hierarchies vs. 1 (ablation) | GeoDecoder · Im2GPS3k | accuracy @ 1 km: +4.5 (9.8 to 14.3) | Improved |
| Train with 7 hierarchies vs. 3 hierarchies (ablation) | GeoDecoder · GWS15k | accuracy @ 1 km: 0.2 vs 0.2 at 1 km; 1 km +0.4 at 5 hierarchies | Mixed |
| Use 16 scene labels instead of 365 (ablation) | GeoDecoder · YFCC26k | accuracy @ 1 km: +0.2 (8.5 to 8.7) | Improved |
| Proposed hierarchical scene-query geo-decoder vs. prior SOTA Translocator | Swin-based GeoDecoder · YFCC26k | accuracy @ 1 km: +2.9 (7.2 to 10.1) | Improved |
Closest papers by meaning
- GeoRanker: Distance-Aware Ranking for Worldwide Image Geolocalization (NeurIPS 2025 · similarity 0.91)
- Recognition through Reasoning: Reinforcing Image Geo-localization with Large Vision-Language Models (NeurIPS 2025 · similarity 0.90)
- G3: An Effective and Adaptive Framework for Worldwide Geolocalization Using Large Multi-Modality Models (NeurIPS 2024 · similarity 0.89)
- GRE Suite: Geo-localization Inference via Fine-Tuned Vision-Language Models and Enhanced Reasoning Chains (NeurIPS 2025 · similarity 0.89)
- Scaling Image Geo-Localization to Continent Level (NeurIPS 2025 · similarity 0.89)
- GeoRouter: Dynamic Paradigm Routing for Worldwide Image Geolocalization (NeurIPS 2026 · 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.
