Doc-V*:Coarse-to-Fine Interactive Visual Reasoning for Multi-Page Document VQA

TL;DR

The paper introduces Doc-V*, an OCR-free, interactive framework for multi-page document VQA that actively navigates and aggregates evidence, significantly improving accuracy and efficiency over existing methods.

Contribution

It presents a novel agentic approach that combines semantic retrieval, targeted page fetching, and evidence aggregation for better multi-page document reasoning.

Findings

Outperforms open-source baselines on five benchmarks.

Improves out-of-domain performance by up to 47.9%.

Effective evidence aggregation achieved with selective attention.

Abstract

Multi-page Document Visual Question Answering requires reasoning over semantics, layouts, and visual elements in long, visually dense documents. Existing OCR-free methods face a trade-off between capacity and precision: end-to-end models scale poorly with document length, while visual retrieval-based pipelines are brittle and passive. We propose Doc-$V^*$, an \textbf{OCR-free agentic} framework that casts multi-page DocVQA as sequential evidence aggregation. Doc-$V^*$ begins with a thumbnail overview, then actively navigates via semantic retrieval and targeted page fetching, and aggregates evidence in a structured working memory for grounded reasoning. Trained by imitation learning from expert trajectories and further optimized with Group Relative Policy Optimization, Doc-$V^*$ balances answer accuracy with evidence-seeking efficiency. Across five benchmarks, Doc-$V^*$ outperforms open-source baselines and approaches proprietary models, improving out-of-domain performance by up to \textbf{47.9\%} over RAG baseline. Other results reveal effective evidence aggregation with selective attention, not increased input pages.