Cascaded Diffusion Models for High Fidelity Image Generation

TL;DR

This paper introduces cascaded diffusion models that generate high-quality images by progressively increasing resolution, with conditioning augmentation crucial for preventing errors and achieving state-of-the-art results on ImageNet.

Contribution

The paper presents a novel cascaded diffusion framework with conditioning augmentation, enabling high-fidelity image generation without auxiliary classifiers, outperforming prior models on ImageNet.

Findings

Achieved FID scores of 1.48, 3.52, and 4.88 at 64x64, 128x128, and 256x256 resolutions.

Outperformed BigGAN-deep in sample quality metrics.

Achieved top-1 and top-5 classification accuracy scores of 63.02% and 84.06% at 256x256.

Abstract

We show that cascaded diffusion models are capable of generating high fidelity images on the class-conditional ImageNet generation benchmark, without any assistance from auxiliary image classifiers to boost sample quality. A cascaded diffusion model comprises a pipeline of multiple diffusion models that generate images of increasing resolution, beginning with a standard diffusion model at the lowest resolution, followed by one or more super-resolution diffusion models that successively upsample the image and add higher resolution details. We find that the sample quality of a cascading pipeline relies crucially on conditioning augmentation, our proposed method of data augmentation of the lower resolution conditioning inputs to the super-resolution models. Our experiments show that conditioning augmentation prevents compounding error during sampling in a cascaded model, helping us to train cascading pipelines achieving FID scores of 1.48 at 64x64, 3.52 at 128x128 and 4.88 at 256x256 resolutions, outperforming BigGAN-deep, and classification accuracy scores of 63.02% (top-1) and 84.06% (top-5) at 256x256, outperforming VQ-VAE-2.