Reproduce “CoInD: Enabling Logical Compositions in Diffusion 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
Sachit Gaudi, Gautam Sreekumar, Vishnu Boddeti
CoInD trains a diffusion model to compose labeled attributes by penalizing disagreement between joint and individual attribute guidance. On Colored MNIST, it improves generation of unseen digit–color combinations under partial training support: AND conformity score is 52.38%, versus 7.40% for Composed GLIDE. This remains imperfect, and the method applies to predefined attribute sets, not arbitrary concepts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.