Reproduce “Loud and Clear: Making Visual Evidence Count in Multimodal Document RAG” 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
Eunchan Jo, Seounghyun Jang, Sangwoo Mo
LoC presents retrieved document pages to a vision-language model with selected OCR text alongside page images, preserving visual context. In an oracle-page SciEGQA test, accuracy rose from 65.74% at a 1K visual-token target to 80.96% at 32K. With pages and generator fixed, LoC reached 78.07% versus 69.19% for ColQwen3; its gains across three benchmarks had paired query-bootstrap intervals excluding zero.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.