GeoAgent: Learning to Geolocate Everywhere with Reinforced Geographic Characteristics

TL;DR

GeoAgent is a novel reinforcement learning model that leverages expert-annotated geographic data and specialized rewards to improve fine-grained geolocation reasoning, outperforming existing methods and aligning closely with human reasoning.

Contribution

The paper introduces GeoAgent, a reinforcement learning approach with new geographic rewards and a specialized dataset, enhancing geolocation accuracy and interpretability.

Findings

GeoAgent outperforms existing methods in geolocation tasks.

It generates reasoning aligned with human geographic understanding.

The approach improves reasoning consistency and accuracy across multiple granularities.

Abstract

This paper presents GeoAgent, a model capable of reasoning closely with humans and deriving fine-grained address conclusions. Previous RL-based methods have achieved breakthroughs in performance and interpretability but still remain concerns because of their reliance on AI-generated chain-of-thought (CoT) data and training strategies, which conflict with geographic characteristics. To address these issues, we first introduce GeoSeek, a new geolocation dataset comprising CoT data annotated by geographic experts and professional players. We further thoroughly explore the inherent characteristics of geographic tasks and propose a geo-similarity reward and a consistency reward assessed by a consistency agent to assist training. This encourages the model to converge towards correct answers from a geographic perspective while ensuring the integrity and consistency of its reasoning process. Experimental results show that GeoAgent outperforms existing methods and a series of general VLLMs across multiple grains, while generating reasoning that closely aligns with humans.