InSight-doc：面向长文档理解的智能体视觉感知

摘要

长文档理解通常需要对大量视觉内容丰富页面进行推理，这导致推理成本高昂且容易发生上下文腐化。在本工作中，我们提出 InSight-doc，一个智能体式视觉感知框架，将视觉分辨率视为一种自适应的推理时间资源。InSight-doc 从低分辨率出发，选择性放大到高分辨率区域以获取更精细的证据，完全无需外部检索器。为训练此类智能体，我们构建了一个包含 17.9K 高质量 SFT 样本的主动感知语料库，其中包含区域级放大轨迹，并配以 19.2K 困难 RL 样本。通过 SFT+RL 训练，InSight-doc-8B 在文档 VQA 基准上将基线提升了 4.3–16.4 个准确率百分点。在长文档上，它使幻觉减少超过 40%，推理延迟降低 41%–68%，同时保持准确率领先。我们的代码、数据集和模型已在 https://github.com/m-Just/InSight-doc 发布。

English

Long-document understanding often requires reasoning over many visually rich pages, making inference costly and prone to context rot. In this work, we propose InSight-doc, an agentic visual perception framework that treats visual resolution as an adaptive reasoning-time resource. InSight-doc starts from low resolution and selectively zooms into high-resolution regions for finer evidence, without relying on any external retriever. To train such an agent, we construct an active-perception corpus of 17.9K high-quality SFT examples with region-level zoom-in trajectories, accompanied by 19.2K hard RL examples. Through SFT+RL, InSight-doc-8B improves the baseline by 4.3--16.4 accuracy points over document VQA benchmarks. On long documents, it reduces hallucination by more than 40% and inference latency by 41%--68% while maintaining an accuracy lead. Our code, datasets, and model are released at https://github.com/m-Just/InSight-doc .