Scalable Diffusion Models with Transformers

ICCV 2023  ·  William Peebles, Saining Xie ·

We explore a new class of diffusion models based on the transformer architecture. We train latent diffusion models of images, replacing the commonly-used U-Net backbone with a transformer that operates on latent patches. We analyze the scalability of our Diffusion Transformers (DiTs) through the lens of forward pass complexity as measured by Gflops. We find that DiTs with higher Gflops -- through increased transformer depth/width or increased number of input tokens -- consistently have lower FID. In addition to possessing good scalability properties, our largest DiT-XL/2 models outperform all prior diffusion models on the class-conditional ImageNet 512x512 and 256x256 benchmarks, achieving a state-of-the-art FID of 2.27 on the latter.

PDF Abstract ICCV 2023 PDF ICCV 2023 Abstract

Datasets


  Add Datasets introduced or used in this paper
Task Dataset Model Metric Name Metric Value Global Rank Result Benchmark
Image Generation ImageNet 256x256 DiT-XL/2 FID 2.27 # 10
Image Generation ImageNet 512x512 DiT-XL/2 FID 3.04 # 13
Inception score 240.82 # 7

Methods