GeoVLM-R1: Reinforcement Fine-Tuning for Improved Remote Sensing Reasoning

2025/09/29 by Mustansar Fiaz, Hiyam Debary, Fiaz, Mustansar +11 · 2 citations
Computer Science · Social Sciences · #Computer Vision and Pattern Recognition (cs.CV) #Data Management and Algorithms #FOS: Computer and information sciences #Geographic Information Systems Studies #Semantic Web and Ontologies

paper · pdf · doi:10.48550/arxiv.2509.25026

openalex publication_date 2025/09/29 · openalex created_date 2025/10/19 · openalex updated_date 2026/07/28

Abstract

Recent advances in reinforcement learning (RL) have delivered strong reasoning capabilities in natural image domains, yet their potential for Earth Observation (EO) remains largely unexplored. EO tasks introduce unique challenges, spanning referred object detection, image or region captioning, change detection, grounding, and temporal analysis, that demand task aware reasoning. We propose a novel post training framework that incorporates task aware rewards to enable effective adaptation of reasoning based RL models to diverse EO tasks. This training strategy enhances reasoning capabilities for remote sensing images, stabilizes optimization, and improves robustness. Extensive experiments across multiple EO benchmarks show consistent performance gains over state of the art generic and specialized vision language models. Code and models will be released publicly at https://mustansarfiaz.github.io/GeoVLM-R1/ .

Citations

• UAV-VL-R1: Generalizing Vision-Language Models via Supervised Fine-Tuning and Multi-Stage GRPO for UAV Visual Reasoning
• R-TPT: Improving Adversarial Robustness of Vision-Language Models through Test-Time Prompt Tuning
• See or Recall: A Sanity Check for the Role of Vision in Solving Visualization Question Answer Tasks with Multimodal LLMs
• SFT or RL? An Early Investigation into Training R1-Like Reasoning Large Vision-Language Models
• VLM-R1: A Stable and Generalizable R1-style Large Vision-Language Model
• Boosting the Generalization and Reasoning of Vision Language Models with Curriculum Reinforcement Learning
• LMM-R1: Empowering 3B LMMs with Strong Reasoning Abilities Through Two-Stage Rule-Based RL
• Vision-R1: Incentivizing Reasoning Capability in Multimodal Large Language Models
• Reward Shaping to Mitigate Reward Hacking in RLHF
• Qwen2.5-VL Technical Report
• GeoPixel: Pixel Grounding Large Multimodal Model in Remote Sensing
• DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning
• EarthDial: Turning Multi-sensory Earth Observations to Interactive Dialogues
• TEOChat: A Large Vision-Language Assistant for Temporal Earth Observation Data
• SkySenseGPT: A Fine-Grained Instruction Tuning Dataset and Model for Remote Sensing Vision-Language Understanding
• Made to Order: Discovering monotonic temporal changes via self-supervised video ordering
• Demonstrating and Reducing Shortcuts in Vision-Language Representation Learning
• BioXP-0.5B: Explainable Medical-AI via RL-GRPO
• LHRS-Bot: Empowering Remote Sensing with VGI-Enhanced Large Multimodal Language Model
• EarthGPT: A Universal Multi-modal Large Language Model for Multi-sensor Image Comprehension in Remote Sensing Domain
• SkyEyeGPT: Unifying Remote Sensing Vision-Language Tasks via Instruction Tuning with Large Language Model
• InternVL: Scaling up Vision Foundation Models and Aligning for Generic Visual-Linguistic Tasks
• Remote Sensing Vision-Language Foundation Models without Annotations via Ground Remote Alignment
• GeoChat: Grounded Large Vision-Language Model for Remote Sensing
• RSGPT: A Remote Sensing Vision Language Model and Benchmark
• RemoteCLIP: A Vision Language Foundation Model for Remote Sensing
• Direct Preference Optimization: Your Language Model is Secretly a Reward Model
• Visual Instruction Tuning
• GPT-4 Technical Report
• Prompt-aligned Gradient for Prompt Tuning
• Supervised Contrastive Learning
• Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks
• Proximal Policy Optimization Algorithms
• High-Dimensional Continuous Control Using Generalized Advantage Estimation
• OpenVLThinker: Complex Vision-Language Reasoning via Iterative SFT-RL Cycles
• EarthMind: Leveraging Cross-Sensor Data for Advanced Earth Observation Interpretation with a Unified Multimodal LLM
• Reason-RFT: Reinforcement Fine-Tuning for Visual Reasoning of Vision Language Models

Cited by

• An Efficient and Effective Encoder Model for Vision and Language Tasks in the Remote Sensing Domain
• GeoZero: Incentivizing Reasoning from Scratch on Geospatial Scenes

Related

• More with Less: a Large Scale Remote Sensing VLM with a Simple Recipe
• RemoteReasoner: Towards Unifying Geospatial Reasoning Workflow
• Few-Shot Vision-Language Reasoning for Satellite Imagery via Verifiable Rewards
• Unlocking Zero-Shot Geospatial Reasoning via Indirect Rewards
• Bridging Semantics and Geometry: A Decoupled LVLM-SAM Framework for Reasoning Segmentation in Optical Remote Sensing