Reproduce “Efficient Hybrid Distillation: Synergizing Score and Adversarial Objectives for One-Step Diffusion” 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
Fei Peng, Junqiang Wu, Haoxian Tan, Jun Zhou, Jie Hu, Xiaoming Wei, Jie Guo, Xiu Li
EASD trains diffusion students for one-step image generation by biasing score-distillation sampling toward higher noise and adversarial-distillation sampling toward lower noise, while retaining coverage across noise levels. The authors’ ablations support these complementary regimes. On ImageNet at 256×256, the full system reports FID 1.39 after 60 epochs with one inference evaluation; this result does not isolate sampling’s effect.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.