Reproduce “DMAD: Distribution Matching as Adversarial Distillation for Fast Visual 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

Zhengming Yu, Junkun Yuan, Haotian Yang, Gordon Guocheng Qian, Yizhi Wang, Angtian Wang, Yiding Yang, Bo Liu and 3 more

DMAD replaces separate score estimation with two classifier heads that compare student outputs with teacher-generated samples and real data. Their logits directly guide a few-step generator, while a measured real-head gap adjusts teacher influence across noise levels. On zero-shot COCO, four-step SDXL achieves FID 14.47, versus 19.32 for DMD2; this result is specific to the reported benchmark.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.