Reproduce “PDMD: Projected Distribution Matching Distillation for Video Diffusion Models” 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
Zimo Wang, Junkun Yuan, Angtian Wang, Haotian Yang, Canyu Zhang, Siyuan Yuan, Xingchang Huang, Bo Liu and 4 more
PDMD stabilizes four-step video diffusion distillation by projecting DMD’s student update away from the residual between a student endpoint and the critic’s one-step endpoint. The projection uses existing predictions and adds no model pass, loss, or training stage. On Wan2.1-T2V-1.3B, PDMD scores 83.73 versus 82.70 for matched DMD at 4 NFE, but convergence and diversity are not guaranteed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.