Denoising Diffusion Probabilistic Models

Code

• github.com/hojonathanho/diffusion
  OfficialIn papertf★ 5,105
• github.com/labmlai/annotated_deep_learning_paper_implementations
  pytorch★ 66,103
• github.com/yang-song/score_sde_pytorch
  pytorch★ 2,088
• github.com/yang-song/score_sde
  pytorch★ 1,810
• github.com/Janspiry/Palette-Image-to-Image-Diffusion-Models
  pytorch★ 1,807
• github.com/glouppe/info8010-deep-learning
  pytorch★ 1,279
• github.com/arpitbansal297/cold-diffusion-models
  pytorch★ 1,120
• github.com/lmnt-com/diffwave
  pytorch★ 889
• github.com/teapearce/conditional_diffusion_mnist
  pytorch★ 825
• github.com/teticio/audio-diffusion
  pytorch★ 789

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

Tasks

Benchmark Results

Reproductions