Search papers, labs, and topics across Lattice.
This paper introduces Optimizing Your Sampling (OYS), a novel approach that utilizes Bayesian optimization to directly select sampling timesteps for diffusion models, thereby enhancing computational efficiency during generation. By treating timestep selection as a black-box optimization problem, OYS significantly improves sample quality in text-to-image generation and other image tasks compared to existing methods. The results show that a 5-step OYS schedule can maintain 89%-94% of the quality of a traditional 50-step schedule while achieving a 10x reduction in inference cost.
A 5-step sampling schedule can deliver nearly the same quality as a 50-step one, slashing inference costs by 90%.
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.
CMU ML