Reproduce “Pay More Attention To Text In High-Resolution MLLMs” 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
Zhongkuan Mao, Wenzhuo Zhao, Xianjie Liu, Yidong Wang, Zhao Gao, Ronghao Xian, Yao Jiang, Yi Zhang and 2 more
EviSpec rewrites questions into target, relation, and option-free search instructions. A frozen multimodal model uses them to select compact evidence views, then answers with the original image and question. It requires no training, detector, or model modification. On V*Bench, Qwen2.5-VL-7B reached 91.6% versus 79.8% with geometry-matched random evidence; this supports relevant selection, not a sole cause or faster inference.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.