Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Vision-Language Reasoning for Geolocalization: A Reinforcement Learning Approach

Biao Wu, Meng Fang, Ling Chen, Ke Xu, Tao Cheng, Jun Wang

TL;DR

This paper addresses global image geolocalization by removing retrieval dependencies and instead leveraging a reasoning-driven framework. It introduces Chain-of-Region reasoning (CoR) and a reinforcement-learning stage using Group Relative Policy Optimization (GRPO) with verifiable rewards based on spatial distance and output format, enabling direct optimization of geographic accuracy. A large-scale synthetic reasoning dataset (MP16-Rand-500K) supports structured supervision, and a hard subset (MP16-Hard-200K) enhances generalization to challenging regions. Experiments on IM2GPS3K and YFCC4K show that Geo-R achieves competitive, retrieval-free performance with strong interpretability and cross-domain robustness, approaching retrieval-based systems. This approach lays the groundwork for scalable, explainable geographic agents that integrate vision-language reasoning with spatial supervision.

Abstract

Recent advances in vision-language models have opened up new possibilities for reasoning-driven image geolocalization. However, existing approaches often rely on synthetic reasoning annotations or external image retrieval, which can limit interpretability and generalizability. In this paper, we present Geo-R, a retrieval-free framework that uncovers structured reasoning paths from existing ground-truth coordinates and optimizes geolocation accuracy via reinforcement learning. We propose the Chain of Region, a rule-based hierarchical reasoning paradigm that generates precise, interpretable supervision by mapping GPS coordinates to geographic entities (e.g., country, province, city) without relying on model-generated or synthetic labels. Building on this, we introduce a lightweight reinforcement learning strategy with coordinate-aligned rewards based on Haversine distance, enabling the model to refine predictions through spatially meaningful feedback. Our approach bridges structured geographic reasoning with direct spatial supervision, yielding improved localization accuracy, stronger generalization, and more transparent inference. Experimental results across multiple benchmarks confirm the effectiveness of Geo-R, establishing a new retrieval-free paradigm for scalable and interpretable image geolocalization. To facilitate further research and ensure reproducibility, both the model and code will be made publicly available.

Vision-Language Reasoning for Geolocalization: A Reinforcement Learning Approach

TL;DR

This paper addresses global image geolocalization by removing retrieval dependencies and instead leveraging a reasoning-driven framework. It introduces Chain-of-Region reasoning (CoR) and a reinforcement-learning stage using Group Relative Policy Optimization (GRPO) with verifiable rewards based on spatial distance and output format, enabling direct optimization of geographic accuracy. A large-scale synthetic reasoning dataset (MP16-Rand-500K) supports structured supervision, and a hard subset (MP16-Hard-200K) enhances generalization to challenging regions. Experiments on IM2GPS3K and YFCC4K show that Geo-R achieves competitive, retrieval-free performance with strong interpretability and cross-domain robustness, approaching retrieval-based systems. This approach lays the groundwork for scalable, explainable geographic agents that integrate vision-language reasoning with spatial supervision.

Abstract

Recent advances in vision-language models have opened up new possibilities for reasoning-driven image geolocalization. However, existing approaches often rely on synthetic reasoning annotations or external image retrieval, which can limit interpretability and generalizability. In this paper, we present Geo-R, a retrieval-free framework that uncovers structured reasoning paths from existing ground-truth coordinates and optimizes geolocation accuracy via reinforcement learning. We propose the Chain of Region, a rule-based hierarchical reasoning paradigm that generates precise, interpretable supervision by mapping GPS coordinates to geographic entities (e.g., country, province, city) without relying on model-generated or synthetic labels. Building on this, we introduce a lightweight reinforcement learning strategy with coordinate-aligned rewards based on Haversine distance, enabling the model to refine predictions through spatially meaningful feedback. Our approach bridges structured geographic reasoning with direct spatial supervision, yielding improved localization accuracy, stronger generalization, and more transparent inference. Experimental results across multiple benchmarks confirm the effectiveness of Geo-R, establishing a new retrieval-free paradigm for scalable and interpretable image geolocalization. To facilitate further research and ensure reproducibility, both the model and code will be made publicly available.
Paper Structure (25 sections, 7 equations, 1 figure, 5 tables)

This paper contains 25 sections, 7 equations, 1 figure, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Image Geolocalization
  4. Verifiable Rewards for VLMs
  5. Methodology
  6. Synthesizing Geographic Reasoning Data at Scale
  7. Reinforcement Learning for Geographic Reasoning
  8. Distance-Based Reward
  9. Format-Based Reward
  10. Diverse Geographic Reasoning Data Selection
  11. Experiments
  12. Datasets
  13. Training Data
  14. Evaluation Benchmarks
  15. Evaluation Metrics
  16. ...and 10 more sections

Figures (1)

  • Figure 1: Overview of Geo-R for image geolocation. Given a query image, the agent generates a region-level reasoning chain grounded in geographic knowledge, followed by coordinate prediction. The framework integrates structured prompting, multimodal understanding, and reinforcement optimization to enhance both interpretability and spatial accuracy.