Scalable Diffusion Models with State Space Backbone

TL;DR

This paper introduces Diffusion State Space Models (DiS), a novel diffusion architecture using state space backbones that effectively handle long-range dependencies, achieving competitive image generation performance with improved scalability and reduced computational costs.

Contribution

The paper proposes DiS, a new diffusion model architecture based on state space models, demonstrating comparable or superior performance to traditional U-Net models with better scalability and efficiency.

Findings

DiS achieves competitive image generation quality on ImageNet benchmarks.

Higher Gflops in DiS correlate with lower FID scores.

Latent space DiS models reduce computational load while maintaining performance.

Abstract

This paper presents a new exploration into a category of diffusion models built upon state space architecture. We endeavor to train diffusion models for image data, wherein the traditional U-Net backbone is supplanted by a state space backbone, functioning on raw patches or latent space. Given its notable efficacy in accommodating long-range dependencies, Diffusion State Space Models (DiS) are distinguished by treating all inputs including time, condition, and noisy image patches as tokens. Our assessment of DiS encompasses both unconditional and class-conditional image generation scenarios, revealing that DiS exhibits comparable, if not superior, performance to CNN-based or Transformer-based U-Net architectures of commensurate size. Furthermore, we analyze the scalability of DiS, gauged by the forward pass complexity quantified in Gflops. DiS models with higher Gflops, achieved through augmentation of depth/width or augmentation of input tokens, consistently demonstrate lower FID. In addition to demonstrating commendable scalability characteristics, DiS-H/2 models in latent space achieve performance levels akin to prior diffusion models on class-conditional ImageNet benchmarks at the resolution of 256$\times$256 and 512$\times$512, while significantly reducing the computational burden. The code and models are available at: https://github.com/feizc/DiS.