Abstract
We present high quality image synthesis results using diffusion probabilisticmodels, a class of latent variable models inspired by considerations fromnonequilibrium thermodynamics. Our best results are obtained by training on aweighted variational bound designed according to a novel connection betweendiffusion probabilistic models and denoising score matching with Langevindynamics, and our models naturally admit a progressive lossy decompressionscheme that can be interpreted as a generalization of autoregressive decoding.On the unconditional CIFAR10 dataset, we obtain an Inception score of 9.46 anda state-of-the-art FID score of 3.17. On 256x256 LSUN, we obtain sample qualitysimilar to ProgressiveGAN. Our implementation is available athttps://github.com/hojonathanho/diffusion