Elucidating the Design Space of Diffusion-Based Generative Models
This work addresses the complexity in diffusion models for researchers and practitioners, offering modular improvements that enhance efficiency and quality, though it is incremental in refining existing methods.
The paper tackles the convoluted theory and practice of diffusion-based generative models by presenting a clear design space that separates concrete design choices, leading to a new state-of-the-art FID of 1.79 for CIFAR-10 in class-conditional settings and 1.36 for ImageNet-64 after re-training, with faster sampling at 35 network evaluations per image.
We argue that the theory and practice of diffusion-based generative models are currently unnecessarily convoluted and seek to remedy the situation by presenting a design space that clearly separates the concrete design choices. This lets us identify several changes to both the sampling and training processes, as well as preconditioning of the score networks. Together, our improvements yield new state-of-the-art FID of 1.79 for CIFAR-10 in a class-conditional setting and 1.97 in an unconditional setting, with much faster sampling (35 network evaluations per image) than prior designs. To further demonstrate their modular nature, we show that our design changes dramatically improve both the efficiency and quality obtainable with pre-trained score networks from previous work, including improving the FID of a previously trained ImageNet-64 model from 2.07 to near-SOTA 1.55, and after re-training with our proposed improvements to a new SOTA of 1.36.