像苏格拉底一样提问：Socrates 帮助视觉语言模型理解遥感图像
Asking like Socrates: Socrates helps VLMs understand remote sensing images
提出「证据链式思维」RS-EoT 范式与 SocraticAgent 自博弈多智能体合成推理轨迹，并用两阶段渐进式强化学习（先细粒度视觉定位、后 VQA）训练 VLM；对做视觉推理、视觉定位与遥感 VQA 的研究者，这是一套可复用的「推理—回看图像」迭代训练流程，可直接迁移到高分辨率或大场景图像理解任务。
原文依据 we propose SocraticAgent, a self-play multi-agent system that synthesizes reasoning traces via alternating cycles of reasoning and visual inspection
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论近期受 DeepSeek-R1 启发的多模态推理模型显著推进了视觉-语言系统的发展。然而在遥感任务中，我们观察到普遍存在的伪推理现象：模型只是在叙述推理过程，而非基于视觉证据真正推理出正确答案。我们将此归因于「一瞥效应」（Glance Effect），即对大尺度遥感影像仅做一次粗粒度感知，导致理解不完整，并依赖语言层面的自洽性而非视觉证据进行推理。为解决这一问题，我们提出 RS-EoT（遥感证据链式思维），一种语言驱动、迭代式的视觉证据搜寻范式。为灌输这一范式，我们提出 SocraticAgent，一个自博弈多智能体系统，通过推理与视觉检查交替循环来合成推理轨迹。为增强并泛化这些模式，我们提出两阶段渐进式强化学习策略：首先在细粒度视觉定位任务上进行强化学习以增强 RS-EoT 能力，随后在遥感视觉问答任务上进行强化学习以泛化到更广泛的理解场景。实验表明，RS-EoT 在多个遥感视觉问答与视觉定位基准上取得最优性能。分析揭示了清晰的推理与证据搜寻迭代循环，证实 RS-EoT 缓解了一瞥效应，并实现了真正基于证据的推理。我们的代码、数据与模型已发布于 https://geox-lab.github.io/Asking_like_Socrates/。
英文原文摘要 1449 字符
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/.