Reproduce “Data-free Multistep Diffusion Distillation” 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

Ngoc Duy Tran, Kien Do, Duc Thanh Nguyen, Truyen Tran

Progressive Multistep Data-free Distillation (PMDD) trains a diffusion image generator in stages, adding denoising steps while preserving earlier-step behavior through frozen previous-stage samples and an unforgetting loss. On 30,000 MS-COCO 2014 validation prompts, three-step Stable Diffusion v1.5 scores FID 7.76 versus 9.02 for data-free DMDv2 at CFG 1.0. The reported scope is few-step sampling, N<4.

Summary by alphaXiv.

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