InSight-doc：面向长文档理解的智能视觉感知框架
原标题：InSight-doc: Agentic Visual Perception for Long-Document Understanding
AI 摘要
InSight-doc 是一种用于长文档理解的智能视觉感知框架，将视觉分辨率视为可自适应调整的推理时资源。它从低分辨率开始，无需外部检索器即可选择性放大高分辨率区域以获取更精细的证据。通过包含17.9K SFT示例和19.2K硬RL示例的训练，InSight-doc-8B在文档VQA基准上提升了4.3-16.4个准确率点，在长文档上将幻觉减少超过40%，推理延迟降低41%-68%，同时保持准确率优势。代码、数据集和模型已在GitHub上发布。
正文节选
InSight-doc: Agentic Visual Perception for Long-Document Understanding Abstract InSight-doc adaptively allocates visual resolution during reasoning to improve long-document understanding while reducing latency and hallucinations. 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