Reproduce “VLMs Read Aloud: The Gap Between Recognizing and Using Visual Text” 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
Zhiyun Shi
The authors introduce ALOUD, which locates evidence in compressed document images, enlarges selected pages, transcribes them, and answers from selected transcript passages. On three long-document question-answering datasets, frozen Qwen3-VL-8B reaches 54.8% accuracy at an effective 7.2× compression, versus 27.7% for uniform rendering and 58.5% from original text; results cover this model and these datasets.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.