Can We Achieve Efficient Diffusion without Self-Attention? Distilling Self-Attention into Convolutions

• URL: http://arxiv.org/abs/2504.21292v1
• Date: Wed, 30 Apr 2025 03:57:28 GMT
• Title: Can We Achieve Efficient Diffusion without Self-Attention? Distilling Self-Attention into Convolutions
• Authors: ZiYi Dong, Chengxing Zhou, Weijian Deng, Pengxu Wei, Xiangyang Ji, Liang Lin,
• Abstract summary: We propose (Delta)ConvFusion to replace conventional self-attention modules with Pyramid Convolution Blocks ((Delta)ConvBlocks)<n>By distilling attention patterns into localized convolutional operations while keeping other components frozen, (Delta)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.
• Score: 94.21989689001848
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: 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 \(\Delta\)ConvFusion to replace conventional self-attention modules with Pyramid Convolution Blocks (\(\Delta\)ConvBlocks).By distilling attention patterns into localized convolutional operations while keeping other components frozen, \(\Delta\)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.

Related papers

• ACDiT: Interpolating Autoregressive Conditional Modeling and Diffusion Transformer [95.80384464922147]
  ACDiT is a blockwise Conditional Diffusion Transformer.<n>It offers a flexible between token-wise autoregression and full-sequence diffusion.<n>We show that ACDiT performs best among all autoregressive baselines on image and video generation tasks.
  arXiv  Detail & Related papers  (2024-12-10T18:13:20Z)
• Efficient Diffusion Transformer with Step-wise Dynamic Attention Mediators [83.48423407316713]
  We present a novel diffusion transformer framework incorporating an additional set of mediator tokens to engage with queries and keys separately. Our model initiates the denoising process with a precise, non-ambiguous stage and gradually transitions to a phase enriched with detail. Our method achieves a state-of-the-art FID score of 2.01 when integrated with the recent work SiT.
  arXiv  Detail & Related papers  (2024-08-11T07:01:39Z)
• Multi-Context Dual Hyper-Prior Neural Image Compression [10.349258638494137]
  We propose a Transformer-based nonlinear transform to efficiently capture both local and global information from the input image. We also introduce a novel entropy model that incorporates two different hyperpriors to model cross-channel and spatial dependencies of the latent representation. Our experiments show that our proposed framework performs better than the state-of-the-art methods in terms of rate-distortion performance.
  arXiv  Detail & Related papers  (2023-09-19T17:44:44Z)
• TGFuse: An Infrared and Visible Image Fusion Approach Based on Transformer and Generative Adversarial Network [15.541268697843037]
  We propose an infrared and visible image fusion algorithm based on a lightweight transformer module and adversarial learning. Inspired by the global interaction power, we use the transformer technique to learn the effective global fusion relations. The experimental performance demonstrates the effectiveness of the proposed modules, with superior improvement against the state-of-the-art.
  arXiv  Detail & Related papers  (2022-01-25T07:43:30Z)
• CSformer: Bridging Convolution and Transformer for Compressive Sensing [65.22377493627687]
  This paper proposes a hybrid framework that integrates the advantages of leveraging detailed spatial information from CNN and the global context provided by transformer for enhanced representation learning. The proposed approach is an end-to-end compressive image sensing method, composed of adaptive sampling and recovery. The experimental results demonstrate the effectiveness of the dedicated transformer-based architecture for compressive sensing.
  arXiv  Detail & Related papers  (2021-12-31T04:37:11Z)
• Improved Transformer for High-Resolution GANs [69.42469272015481]
  We introduce two key ingredients to Transformer to address this challenge. We show in the experiments that the proposed HiT achieves state-of-the-art FID scores of 31.87 and 2.95 on unconditional ImageNet $128 times 128$ and FFHQ $256 times 256$, respectively.
  arXiv  Detail & Related papers  (2021-06-14T17:39:49Z)
• X-volution: On the unification of convolution and self-attention [52.80459687846842]
  We propose a multi-branch elementary module composed of both convolution and self-attention operation. The proposed X-volution achieves highly competitive visual understanding improvements.
  arXiv  Detail & Related papers  (2021-06-04T04:32:02Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.