Improved Denoising Diffusion Probabilistic Models

Code

• github.com/openai/improved-diffusion
  OfficialIn paperpytorch★ 3,811
• github.com/neonbjb/tortoise-tts
  pytorch★ 14,824
• github.com/open-mmlab/mmgeneration
  pytorch★ 2,016
• github.com/luping-liu/PNDM
  pytorch★ 355
• github.com/plai-group/flexible-video-diffusion-modeling
  pytorch★ 120
• github.com/intuitive-robots/beso
  pytorch★ 89
• github.com/Michedev/DDPM-Pytorch
  pytorch★ 88
• github.com/albarji/toy-diffusion
  pytorch★ 47
• github.com/rowerliu/add
  pytorch★ 12
• github.com/rowerliu/adbd
  pytorch★ 12

Abstract

Denoising diffusion probabilistic models (DDPM) are a class of generative models which have recently been shown to produce excellent samples. We show that with a few simple modifications, DDPMs can also achieve competitive log-likelihoods while maintaining high sample quality. Additionally, we find that learning variances of the reverse diffusion process allows sampling with an order of magnitude fewer forward passes with a negligible difference in sample quality, which is important for the practical deployment of these models. We additionally use precision and recall to compare how well DDPMs and GANs cover the target distribution. Finally, we show that the sample quality and likelihood of these models scale smoothly with model capacity and training compute, making them easily scalable. We release our code at https://github.com/openai/improved-diffusion

Tasks

Benchmark Results

Reproductions