GeoAgent: Learning to Geolocate Everywhere with Reinforced Geographic Characteristics

Researchers introduced GeoAgent, a model that outperforms existing methods and general vision-language models in fine-grained geolocation tasks. The key innovation is GeoSeek—a dataset annotated by geographic experts and professional players—which provides high-quality chain-of-thought data. The model uses geo-similarity and consistency rewards during training, ensuring reasoning aligns with geographic principles. Experimental results show GeoAgent generates human-like reasoning while achieving superior accuracy across multiple granularities.