Guardado en:
Detalles Bibliográficos
Autores principales: Dong, ZiYi, Zhou, Chengxing, Deng, Weijian, Wei, Pengxu, Ji, Xiangyang, Lin, Liang
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:https://arxiv.org/abs/2504.21292
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866909597616308224
author Dong, ZiYi
Zhou, Chengxing
Deng, Weijian
Wei, Pengxu
Ji, Xiangyang
Lin, Liang
author_facet Dong, ZiYi
Zhou, Chengxing
Deng, Weijian
Wei, Pengxu
Ji, Xiangyang
Lin, Liang
contents Contemporary diffusion models built upon U-Net or Diffusion Transformer (DiT) architectures have revolutionized image generation through transformer-based attention mechanisms. The prevailing paradigm has commonly employed self-attention with quadratic computational complexity to handle global spatial relationships in complex images, thereby synthesizing high-fidelity images with coherent visual semantics.Contrary to conventional wisdom, our systematic layer-wise analysis reveals an interesting discrepancy: self-attention in pre-trained diffusion models predominantly exhibits localized attention patterns, closely resembling convolutional inductive biases. This suggests that global interactions in self-attention may be less critical than commonly assumed.Driven by this, we propose \(Δ\)ConvFusion to replace conventional self-attention modules with Pyramid Convolution Blocks (\(Δ\)ConvBlocks).By distilling attention patterns into localized convolutional operations while keeping other components frozen, \(Δ\)ConvFusion achieves performance comparable to transformer-based counterparts while reducing computational cost by 6929$\times$ and surpassing LinFusion by 5.42$\times$ in efficiency--all without compromising generative fidelity.
format Preprint
id arxiv_https___arxiv_org_abs_2504_21292
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Can We Achieve Efficient Diffusion without Self-Attention? Distilling Self-Attention into Convolutions
Dong, ZiYi
Zhou, Chengxing
Deng, Weijian
Wei, Pengxu
Ji, Xiangyang
Lin, Liang
Computer Vision and Pattern Recognition
Contemporary diffusion models built upon U-Net or Diffusion Transformer (DiT) architectures have revolutionized image generation through transformer-based attention mechanisms. The prevailing paradigm has commonly employed self-attention with quadratic computational complexity to handle global spatial relationships in complex images, thereby synthesizing high-fidelity images with coherent visual semantics.Contrary to conventional wisdom, our systematic layer-wise analysis reveals an interesting discrepancy: self-attention in pre-trained diffusion models predominantly exhibits localized attention patterns, closely resembling convolutional inductive biases. This suggests that global interactions in self-attention may be less critical than commonly assumed.Driven by this, we propose \(Δ\)ConvFusion to replace conventional self-attention modules with Pyramid Convolution Blocks (\(Δ\)ConvBlocks).By distilling attention patterns into localized convolutional operations while keeping other components frozen, \(Δ\)ConvFusion achieves performance comparable to transformer-based counterparts while reducing computational cost by 6929$\times$ and surpassing LinFusion by 5.42$\times$ in efficiency--all without compromising generative fidelity.
title Can We Achieve Efficient Diffusion without Self-Attention? Distilling Self-Attention into Convolutions
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2504.21292