Reproduce “LEAP: Accelerating ODE-based Diffusion Sampling with Optimal Linear Leaps” 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
Haowei Zheng, Zhenyu Zhou, Jiawei Chen, Can Wang, Defang Chen
LEAP predicts some diffusion sampler outputs from recent history using learned linear rules, alternating predictions with exact model evaluations to limit drift. It fits coefficients from reference trajectories before inference. On SD3.5-Medium at 10 steps, LEAP with DPM++ reports higher HPSv2, ImageReward, and CLIPScore than DPM++ alone; these results apply only to the tested model, task, step budget, and comparison.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.