优化你的采样：基于贝叶斯优化的调优扩散采样
Optimize Your Sampling: Tuned Diffusion Sampling with Bayesian Optimization
浏览论文内容
中文总结 AI 辅助
本研究提出OYS方法，将扩散模型采样的时间步长选择作为黑盒优化问题，用贝叶斯优化直接优化目标指标，可提升多类图像生成任务性能，大幅降低推理成本且无需额外训练。
中文摘要 AI 辅助
从扩散模型采样通常需要对大型神经网络进行多次前向传播，这使得生成过程的计算成本很高。尽管已有大量工作聚焦于高效求解器和采样器，但针对采样时间步长本身的选择却鲜有关注。近期一系列工作优化的是从理论推导的样本质量替代物，而非质量指标本身。我们提出了OYS（即Optimize Your Sampling），将时间步长选择视为黑盒优化问题，直接通过贝叶斯优化对目标指标进行优化。在文本到图像生成任务中，OYS的表现优于默认时间步长方案和Align Your Steps的方案；在图像修复及其他图像任务中，OYS也优于默认方案，且在定量评估和人工评估中均表现出色。OYS无需额外训练，甚至适用于蒸馏模型，还能提升Euler、DPM-Solver++等简单及复杂采样器的性能。5步OYS方案可保留50步方案89%-94%的质量，同时将推理成本降低10倍。
英文摘要
Sampling from a diffusion model typically requires many forward passes through a large neural network, making generation computationally expensive. While much work has focused on efficient solvers and samplers, comparatively little attention has been paid to selecting the sampling timesteps themselves. A recent line of work optimizes theoretically derived surrogates for sample quality rather than the quality metric itself. We propose Optimizing Your Sampling (OYS), which instead treats timestep selection as a black-box optimization problem, optimizing the target metric directly with Bayesian optimization. OYS outperforms both the default schedules and those of Align Your Steps on text-to-image generation, and improves over the default schedules on inpainting and other image tasks, in both quantitative and human evaluations. OYS requires no additional training, is applicable even to distilled models, and improves both simple and sophisticated samplers such as Euler and DPM-Solver++. A 5-step OYS schedule retains 89%-94% of the quality of a 50-step schedule while reducing inference cost by 10x.
发表机构
- Cornell University（康奈尔大学）
机构由 AI 辅助整理，请以论文原文为准。
