Relay Diffusion: Unifying diffusion process across resolutions for image synthesis

TL;DR

Relay Diffusion introduces a novel method to unify diffusion processes across different resolutions, enabling seamless high-resolution image synthesis and achieving state-of-the-art results on CelebA-HQ and ImageNet 256x256.

Contribution

It proposes Relay Diffusion Model (RDM), which transfers low-resolution images into high-resolution ones for diffusion, allowing continuous high-res generation without restarting from scratch.

Findings

Achieves state-of-the-art FID on CelebA-HQ

Surpasses previous models on ImageNet 256x256 sFID

Enables seamless diffusion across resolutions

Abstract

Diffusion models achieved great success in image synthesis, but still face challenges in high-resolution generation. Through the lens of discrete cosine transformation, we find the main reason is that \emph{the same noise level on a higher resolution results in a higher Signal-to-Noise Ratio in the frequency domain}. In this work, we present Relay Diffusion Model (RDM), which transfers a low-resolution image or noise into an equivalent high-resolution one for diffusion model via blurring diffusion and block noise. Therefore, the diffusion process can continue seamlessly in any new resolution or model without restarting from pure noise or low-resolution conditioning. RDM achieves state-of-the-art FID on CelebA-HQ and sFID on ImageNet 256$\times$256, surpassing previous works such as ADM, LDM and DiT by a large margin. All the codes and checkpoints are open-sourced at \url{https://github.com/THUDM/RelayDiffusion}.