Reproduce “An End-to-End Latent-Rollout Approach for Pushing Few-Step ImageNet-$256$ Generation to FID $1.11$ without Fréchet Losses” 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
Xiaoran Xu, Yujing Wang
FiST replaces repeated diffusion sampling with a two- or three-stage latent rollout. It first learns transitions from selected states of a pretrained teacher, then refines its final outputs using real-data realism and class objectives. On class-conditional ImageNet at 256×256, three-stage augmented FiST reports FID 1.11 and IS 282, below two-step FACM’s reported FID 1.32; this is not a controlled comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.