Reproduce “Adversarial score matching and improved sampling for 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
Alexia Jolicoeur-Martineau, Rémi Piché-Taillefer, Rémi Tachet des Combes, Ioannis Mitliagkas
The paper introduces consistent annealed sampling (CAS) and final expected denoised sample estimates for score-based image generation. CAS corrects the sampling noise schedule; denoising estimates the clean image from the final noisy sample. On CIFAR-10, these changes lower FID, including a reduction from 14.7 to 12.3 in a controlled comparison, but results do not establish general superiority to GANs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.