Reproduce “DocPilot: Interactive Agent Orchestration and Training for Long Visual Document Understanding” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zhuowen Liang, Xiaotian Lin, Yuyu Luo, Haixun Wang, Nan Tang
DocPilot lets compact models answer long-document questions by navigating pages, inspecting text and visual details, calculating values, and checking answers in a revisable evidence loop. On MMLongBench-Doc, DocPilot-4B reaches 59.1 overall accuracy, versus 54.9 for vanilla Qwen3.5-4B, a 4.2-point gain. Improvement is not uniform: its layout score is 9.8 points below baseline.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.