Reproduce “Focused Omni-Modal GUI Agents” 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
Fuyao Zhang, Wenwen He, Jiaming Zhang, Yurong Hao, Xuan Wang, Guowei Guan, Che Wang, Longtao Huang and 1 more
FocusGUI guides a GUI decoder by scoring observation channels against a reference recomputed at each step, then adding compressed evidence to existing prompt embeddings before action generation. It preserves the full multimodal input and trains on next-action supervision. On OmniGUI, across four frozen backbones, average Type Match rose from 31.0% to 64.2%; this result applies to the benchmark and models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.