GeoRouter: Dynamic Paradigm Routing for Worldwide Image Geolocalization
🤖 AI Summary
This work addresses the challenge of global image geolocation, which demands both fine-grained visual matching and high-level semantic reasoning yet remains difficult for any single paradigm to handle due to substantial visual and geographic diversity. To this end, we propose GeoRouter, a dynamic routing framework built upon large vision-language models (LVLMs) that adaptively selects between retrieval-based and generation-based geolocation pathways according to the input image. Our key contributions include the first dynamic routing mechanism tailored for geolocation, the construction of GeoRouting—the first large-scale dataset for training routing policies—and a distance-aware preference optimization objective. Experiments demonstrate that GeoRouter significantly outperforms existing methods on IM2GPS3k and YFCC4k, confirming its effectiveness and superiority.
Application Category
📝 Abstract
Worldwide image geolocalization aims to predict precise GPS coordinates for images captured anywhere on Earth, which is challenging due to the large visual and geographic diversity. Recent methods mainly follow two paradigms: retrieval-based approaches that match queries against a reference database, and generation-based approaches that directly predict coordinates using Large Vision-Language Models (LVLMs). However, we observe distinct error profiles between them: retrieval excels at fine-grained instance matching, while generation offers robust semantic reasoning. This complementary heterogeneity suggests that no single paradigm is universally superior. To harness this potential, we propose GeoRouter, a dynamic routing framework that adaptively assigns each query to the optimal paradigm. GeoRouter leverages an LVLM backbone to analyze visual content and provide routing decisions. To optimize GeoRouter, we introduce a distance-aware preference objective that converts the distance gap between paradigms into a continuous supervision signal, explicitly reflecting relative performance differences. Furthermore, we construct GeoRouting, the first large-scale dataset tailored for training routing policies with independent paradigm predictions. Extensive experiments on IM2GPS3k and YFCC4k demonstrate that GeoRouter significantly outperforms state-of-the-art baselines.
Problem
Research questions and friction points this paper is trying to address.
worldwide image geolocalization
paradigm routing
retrieval-based
generation-based
geographic diversity
Innovation
Methods, ideas, or system contributions that make the work stand out.
dynamic routing
geolocalization
vision-language models
retrieval-generation fusion
preference optimization
🔎 Similar Papers
No similar papers found.
