GeoZero: Incentivizing Reasoning from Scratch on Geospatial Scenes

TL;DR

GeoZero tackles the challenge of enabling geospatial reasoning in multimodal LLMs without predefined chain-of-thought supervision. It introduces GeoZero-Instruct for supervised fine-tuning and GeoZero-Hard for reinforcement learning, guided by the novel A^2GRPO framework that anchors thinking to answer quality while promoting diverse reasoning. Across SC, VG, VQA, and IC tasks, GeoZero achieves competitive or state-of-the-art performance and demonstrates emergent, scratch-built reasoning trajectories, including clear, interpretable intermediate reasoning. By reducing annotation costs and enabling universal geospatial cognition, GeoZero offers a scalable path toward more transparent and capable geospatial AI systems.

Abstract

Multimodal large language models (MLLMs) have undergone rapid development in advancing geospatial scene understanding. Recent studies have sought to enhance the reasoning capabilities of remote sensing MLLMs, typically through cold-start training with elaborately curated chain-of-thought (CoT) data. However, this approach not only incurs substantial annotation costs but also introduces human biases that may limit the diversity of model reasoning. To address these challenges, we propose GeoZero, a framework that enables MLLMs to perform geospatial reasoning without any predefined CoT supervision. Specifically, we construct two datasets, GeoZero-Instruct and GeoZero-Hard. GeoZero-Instruct allows the model to acquire preliminary geospatial knowledge through supervised fine-tuning, while GeoZero-Hard stimulates deep reasoning during the subsequent reinforcement learning stage. Furthermore, we introduce Answer-Anchored Group Relative Policy Optimization (A$^2$GRPO), where the reasoning process is regularized by the model's own answers, encouraging diverse yet accurate thinking. Extensive experiments on multiple remote sensing vision-language benchmarks demonstrate that GeoZero not only surpasses existing state-of-the-art methods but also fosters universal emergent reasoning capabilities across diverse geospatial tasks. Code,data,and models will be publicly available at https://github.com/MiliLab/GeoZero.

Figures (14)

• Figure 1: Compared with previous methods, GeoZero enables the emergence of reasoning on geospatial scenes without any CoT supervision while maintaining answer correctness.
• Figure 2: Overall framework of GeoZero. We first construct two datasets, GeoZero-Instruct and GeoZero-Hard, for SFT and RL, respectively. The SFT stage leverages remote sensing instruct-following data without CoT to help MLLMs acquire basic geospatial knowledge, while the subsequent RL stage with A$^2$GRPO further enhances the reasoning ability, enabling the model to think from scratch.
• Figure 3: Relationship between thinking and answering, where the statistics are computed from GeoZero’s prediction results on the DIOR-RSVG test set. (Best view in zoom.)
• Figure 4: Visualization of reasoning processes generated by GeoZero across different tasks. Examples are drawn from the test sets of AID, DIOR-RSVG, and RSVQA-HR. Red text highlights the model's reasoning steps leading to the final answers. GT: ground truth.
• Figure 5: Example code for evaluating reasoning activation.