Reproduce “COMPASS: Grounding Composition-Intent Guidance in Unified Multimodal Models” 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
Ziqi Zhou, Weize Quan, Mining Tan, Zhihan Chen, Dandan Zheng, Jingdong Chen, Jun Zhou, Weiming Dong and 1 more
COMPASS represents photographic composition as a learnable token and uses it to guide image generation from a pixelated reference and a text prompt. On Comp-11, it reaches 90.6 mAP versus 58.2 for Ming-Lite-Uni. An independent evaluation finds better composition agreement and prompt alignment than BAGEL, but BAGEL has better FID, and layout transfer is not guaranteed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.