Improved Denoising Diffusion Probabilistic Models

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

References in corpus (14)

• Adam: A Method for Stochastic Optimization
• Deep Residual Learning for Image Recognition
• GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium
• Large Scale GAN Training for High Fidelity Natural Image Synthesis
• Scaling Laws for Neural Language Models
• Deep Unsupervised Learning using Nonequilibrium Thermodynamics
• Generating Long Sequences with Sparse Transformers
• NVAE: A Deep Hierarchical Variational Autoencoder
• Improved Techniques for Training Score-Based Generative Models
• An Empirical Model of Large-Batch Training
• DiffWave: A Versatile Diffusion Model for Audio Synthesis
• Generating High Fidelity Images with Subscale Pixel Networks and Multidimensional Upscaling
• Very Deep VAEs Generalize Autoregressive Models and Can Outperform Them on Images
• Adversarial score matching and improved sampling for image generation

Cited by in corpus (33)

• Diffusion Models Beat GANs on Image Synthesis
• Variational Diffusion Models
• DPM-Solver++: Fast Solver for Guided Sampling of Diffusion Probabilistic Models
• InterGen: Diffusion-based Multi-human Motion Generation under Complex Interactions
• Structured Denoising Diffusion Models in Discrete State-Spaces
• CSDI: Conditional Score-based Diffusion Models for Probabilistic Time Series Imputation
• Deep Generative Modelling: A Comparative Review of VAEs, GANs, Normalizing Flows, Energy-Based and Autoregressive Models
• UNIT-DDPM: UNpaired Image Translation with Denoising Diffusion Probabilistic Models
• Diffusion Schrödinger Bridge with Applications to Score-Based Generative Modeling
• ImageBART: Bidirectional Context with Multinomial Diffusion for Autoregressive Image Synthesis
• Learning to Efficiently Sample from Diffusion Probabilistic Models
• Development of Skip Connection in Deep Neural Networks for Computer Vision and Medical Image Analysis: A Survey
• Maximum Likelihood Training of Score-Based Diffusion Models
• Vector Quantized Diffusion Model for Text-to-Image Synthesis
• Argmax Flows and Multinomial Diffusion: Learning Categorical Distributions
• A Study on Speech Enhancement Based on Diffusion Probabilistic Model
• Enhancing Hyperspectral Images via Diffusion Model and Group-Autoencoder Super-resolution Network
• Gotta Go Fast When Generating Data with Score-Based Models
• How Much is Enough? A Study on Diffusion Times in Score-based Generative Models
• SegDiff: Image Segmentation with Diffusion Probabilistic Models
• Virtual Staining of Label-Free Tissue in Imaging Mass Spectrometry
• Score-based Generative Modeling in Latent Space
• Bilateral Denoising Diffusion Models
• Densely connected normalizing flows
• LaMD: Latent Motion Diffusion for Image-Conditional Video Generation
• Diff-TTS: A Denoising Diffusion Model for Text-to-Speech
• Diffusion Priors In Variational Autoencoders
• Zero-Shot Paragraph-level Handwriting Imitation with Latent Diffusion Models
• Likelihood Training of Schrödinger Bridge using Forward-Backward SDEs Theory
• Density Ratio Estimation via Infinitesimal Classification
• Image Captions are Natural Prompts for Text-to-Image Models
• Diffusion Normalizing Flow
• Lane Graph Extraction from Aerial Imagery via Lane Segmentation Refinement with Diffusion Models