Denoising Diffusion Probabilistic Models

TL;DR

This paper introduces diffusion probabilistic models for high-quality image synthesis, leveraging a novel connection to denoising score matching and Langevin dynamics, achieving state-of-the-art results on CIFAR10 and competitive performance on LSUN.

Contribution

The paper proposes a new class of diffusion probabilistic models with a novel training bound and a progressive decoding scheme, advancing the state-of-the-art in image generation.

Findings

Achieved an Inception score of 9.46 on CIFAR10.

Obtained a FID score of 3.17 on CIFAR10.

Produced image quality comparable to ProgressiveGAN on LSUN.

Abstract

We present high quality image synthesis results using diffusion probabilistic models, a class of latent variable models inspired by considerations from nonequilibrium thermodynamics. Our best results are obtained by training on a weighted variational bound designed according to a novel connection between diffusion probabilistic models and denoising score matching with Langevin dynamics, and our models naturally admit a progressive lossy decompression scheme that can be interpreted as a generalization of autoregressive decoding. On the unconditional CIFAR10 dataset, we obtain an Inception score of 9.46 and a state-of-the-art FID score of 3.17. On 256x256 LSUN, we obtain sample quality similar to ProgressiveGAN. Our implementation is available at https://github.com/hojonathanho/diffusion