Geo-R1: Unlocking VLM Geospatial Reasoning with Cross-View Reinforcement Learning

2025-09-29

AI Summary

Vision-language models (VLMs) lack explicit spatial reasoning capabilities, rely heavily on costly manual annotations, and struggle to integrate visual cues with geographic prior knowledge for geospatial reasoning. Method: We propose Geo-R1—the first inference-centric post-training framework for VLMs—featuring: (1) a novel “reasoning guidance → reasoning enhancement” two-stage paradigm, shifting geospatial modeling from pretraining/supervised fine-tuning to inference-oriented post-training; (2) verifiable and scalable reward signals derived from synthetic chain-of-thought samples and weakly supervised cross-view contrastive proxies; and (3) joint optimization via supervised fine-tuning and GRPO-based reinforcement learning to align cross-modal features and refine reasoning. Results: Geo-R1 achieves state-of-the-art performance across multiple geospatial reasoning benchmarks, significantly improving generalization and reasoning capability in zero-annotation settings. The code and models are publicly released.

Technology Category

• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Computer Vision: Remote Sensing / Geospatial AI
• Natural Language Processing: Language Grounding & Multi-modal NLP

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

We introduce Geo-R1, a reasoning-centric post-training framework that unlocks geospatial reasoning in vision-language models by combining thinking scaffolding and elevating. In the scaffolding stage, Geo-R1 instills a ``geospatial thinking paradigm" via supervised fine-tuning on synthetic chain-of-thought exemplars, enabling models to connect visual cues with geographic priors without costly human reasoning annotations. In the elevating stage, it uses GRPO-based reinforcement learning on a weakly-supervised cross-view pairing proxy. This design supplies a verifiable and scalable reward signal: teaching models to capture and reconcile features across modalities, and harnessing reasoning for accurate prediction. Geo-R1 extends geospatial modeling from domain pretraining / supervised finetuning to reasoning-first post-training, and achieves state-of-the-art performance across various geospatial reasoning benchmarks. Our model is available at https://huggingface.co/miniHui/Geo-R1.

Problem

• Unlocking geospatial reasoning in vision-language models through cross-view reinforcement learning
• Teaching models to connect visual cues with geographic priors without human annotations
• Enhancing cross-modal feature reconciliation for accurate geospatial predictions

Innovation

• Cross-view reinforcement learning for geospatial reasoning
• Synthetic chain-of-thought training without human annotations
• Weakly-supervised cross-view pairing with verifiable reward signals