InSight-doc: Agentic Visual Perception for Long-Document Understanding

2026-08-11

AI Summary

This work addresses the challenges of high inference cost, contextual degradation, and hallucination in long-document visual understanding. The authors propose an agent-based visual perception framework that treats visual resolution as an adaptively allocable resource, starting from low-resolution inputs and dynamically zooming into critical regions to acquire fine-grained evidence—without relying on external retrievers. They innovatively construct active perception–based training data and train an 8B-parameter model using a combination of supervised fine-tuning and reinforcement learning (SFT+RL), leveraging 17.9K region-level zoom trajectories and 19.2K hard examples. On document VQA benchmarks, the method achieves accuracy gains of 4.3–16.4 percentage points, reduces hallucination by over 40%, and cuts inference latency by 41%–68%, all while preserving strong accuracy performance.

Abstract

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 .

Problem

• long-document understanding
• visual perception
• context rot
• inference cost
• document VQA

Innovation

• agentic visual perception
• adaptive resolution
• active perception
• long-document understanding
• SFT+RL training