arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2412.10786cs.CVcs.AI
收藏

优化扩散概率模型的少步采样器

Optimizing Few-Step Sampler for Diffusion Probabilistic Model

Peking University(北京大学)

Jen-Yuan Huang

AI总结:

针对扩散模型推理成本高的问题,提出通过优化采样调度和微调模型的两阶段交替算法,减少少步采样的离散化误差,在ImageNet64上持续提升性能。

摘要:

扩散概率模型(DPMs)在生成高质量、多样化的图像方面展现了卓越的能力,但其实际应用受到推理过程中密集计算成本的阻碍。DPM的生成过程需要求解概率流常微分方程(PF-ODE),这涉及将积分域离散化为区间以进行数值近似。这对应于扩散ODE求解器的采样调度,我们注意到一阶求解器的解可以表示为所有计划时间步上模型输出的凸组合。我们推导了采样调度离散化误差的上界,该上界可以通过蒙特卡洛估计高效优化。基于这些理论结果,我们提出了一种两阶段交替优化算法。在第一阶段,针对预训练的DPM优化采样调度;在第二阶段,在选定的时间步上进一步微调DPM。在ImageNet64数据集上对预训练DPM的实验表明,所提出的方法在各种采样步数下均持续改善了基线性能。

英文摘要:

Diffusion Probabilistic Models (DPMs) have demonstrated exceptional capability of generating high-quality and diverse images, but their practical application is hindered by the intensive computational cost during inference. The DPM generation process requires solving a Probability-Flow Ordinary Differential Equation (PF-ODE), which involves discretizing the integration domain into intervals for numerical approximation. This corresponds to the sampling schedule of a diffusion ODE solver, and we notice the solution from a first-order solver can be expressed as a convex combination of model outputs at all scheduled time-steps. We derive an upper bound for the discretization error of the sampling schedule, which can be efficiently optimized with Monte-Carlo estimation. Building on these theoretical results, we purpose a two-phase alternating optimization algorithm. In Phase-1, the sampling schedule is optimized for the pre-trained DPM; in Phase-2, the DPM further tuned on the selected time-steps. Experiments on a pre-trained DPM for ImageNet64 dataset demonstrate the purposed method consistently improves the baseline across various number of sampling steps.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑