Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

U-DiTs: Downsample Tokens in U-Shaped Diffusion Transformers

About

Diffusion Transformers (DiTs) introduce the transformer architecture to diffusion tasks for latent-space image generation. With an isotropic architecture that chains a series of transformer blocks, DiTs demonstrate competitive performance and good scalability; but meanwhile, the abandonment of U-Net by DiTs and their following improvements is worth rethinking. To this end, we conduct a simple toy experiment by comparing a U-Net architectured DiT with an isotropic one. It turns out that the U-Net architecture only gain a slight advantage amid the U-Net inductive bias, indicating potential redundancies within the U-Net-style DiT. Inspired by the discovery that U-Net backbone features are low-frequency-dominated, we perform token downsampling on the query-key-value tuple for self-attention that bring further improvements despite a considerable amount of reduction in computation. Based on self-attention with downsampled tokens, we propose a series of U-shaped DiTs (U-DiTs) in the paper and conduct extensive experiments to demonstrate the extraordinary performance of U-DiT models. The proposed U-DiT could outperform DiT-XL/2 with only 1/6 of its computation cost. Codes are available at https://github.com/YuchuanTian/U-DiT.

Yuchuan Tian, Zhijun Tu, Hanting Chen, Jie Hu, Chao Xu, Yunhe Wang• 2024

Related benchmarks

TaskDatasetResultRank
Semantic segmentationADE20K (val)
mIoU46.91
2731
Class-conditional Image GenerationImageNet 256x256 (train)
IS246
305
Class-conditional Image GenerationImageNet 256x256 (val)
FID3.37
293
Image GenerationImageNet 256x256
FID10.08
243
Image GenerationImageNet 256x256 (train)
FID10.08
91
Image GenerationImageNet 512x512 (test)
FID15.39
57
Image ClassificationImageNet-1k (val)
Top-1 Accuracy85.13
32
Image SynthesisImageNet 512x512
FID15.39
7
Image GenerationImageNet 512x512 (train)
FID15.39
6
Showing 9 of 9 rows

Other info

Code

Follow for update