Asking like Socrates: Socrates helps VLMs understand remote sensing images

2025-11-27

AI Summary

A pervasive “pseudo-reasoning” problem exists in remote sensing multimodal reasoning: models generate answers based on linguistic self-consistency rather than visual evidence, primarily due to single-pass coarse perception (the “Glance Effect”) of high-resolution remote sensing imagery. To address this, we propose RS-EoT—a language-driven iterative visual evidence search paradigm—and introduce SocraticAgent, a Socratic-style multi-agent framework. SocraticAgent integrates self-play, fine-grained object localization via reinforcement learning, and a two-stage progressive vision-language pretraining and VQA fine-tuning pipeline, enabling an interpretable evidence-reasoning closed loop. RS-EoT significantly mitigates the Glance Effect, achieving state-of-the-art performance across multiple remote sensing visual question answering and referring expression grounding benchmarks. Empirical evaluation confirms its capacity for transparent, traceable visual evidence interaction and stepwise reasoning.

Technology Category

• Computer Vision: Visual Reasoning & Symbolic Representations
• Intelligent Robots: Multimodal Perception & Sensor Fusion
• Knowledge Representation and Reasoning: Common-Sense Reasoning

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Responsible Web: Machine-in-the-loop, human agency and autonomy

Abstract

Recent multimodal reasoning models, inspired by DeepSeek-R1, have significantly advanced vision-language systems. However, in remote sensing (RS) tasks, we observe widespread pseudo reasoning: models narrate the process of reasoning rather than genuinely reason toward the correct answer based on visual evidence. We attribute this to the Glance Effect, where a single, coarse perception of large-scale RS imagery results in incomplete understanding and reasoning based on linguistic self-consistency instead of visual evidence. To address this, we propose RS-EoT (Remote Sensing Evidence-of-Thought), a language-driven, iterative visual evidence-seeking paradigm. To instill this paradigm, we propose SocraticAgent, a self-play multi-agent system that synthesizes reasoning traces via alternating cycles of reasoning and visual inspection. To enhance and generalize these patterns, we propose a two-stage progressive RL strategy: first, RL on fine-grained Grounding tasks to enhance RS-EoT capabilities, followed by RL on RS VQA to generalize to broader understanding scenarios. Experiments show RS-EoT achieves state-of-the-art performance on multiple RS VQA and grounding benchmarks. Analyses reveal clear iterative cycles of reasoning and evidence seeking, confirming RS-EoT mitigates the Glance Effect and enables genuine evidence-grounded reasoning. Our code, data, and models are available at https://geox-lab.github.io/Asking_like_Socrates

Problem

• Addresses pseudo reasoning in remote sensing vision-language models
• Mitigates the Glance Effect for evidence-based image understanding
• Enables iterative visual evidence-seeking in remote sensing tasks

Innovation

• Language-driven iterative visual evidence-seeking paradigm
• Self-play multi-agent system synthesizing reasoning traces
• Two-stage progressive reinforcement learning strategy