Reproduce “CCG Diffusion: Combinatory Categorial Grammar-Driven Compositional and Syntactically Disambiguated Text-to-Image Generation” 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
Ryosuke Yamaki, Daichi Mochihashi, Nobutaka Shimada, Seitaro Shinagawa, Tadahiro Taniguchi
CCG Diffusion uses Combinatory Categorial Grammar parses to extract word relations, rewrite prompts, and guide attention during generation, without retraining. It supports compositional control and user-selected prepositional-phrase attachment. On Sana 0.6B it scores 0.4525 on T2I-CompBench++’s 2D-spatial category, versus 0.3311 for Attend-and-Excite; it also reaches 82.7% and 78.6% on PP-attachment evaluation, but these results do not establish general ambiguity resolution.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.