Can We Achieve Efficient Diffusion without Self-Attention? Distilling Self-Attention into Convolutions
This work addresses efficiency issues in image generation for AI researchers and practitioners, though it is incremental as it builds on existing diffusion model architectures.
The paper tackled the problem of high computational cost in diffusion models by replacing self-attention modules with convolutional blocks, achieving comparable performance while reducing computational cost by 6929× and surpassing LinFusion by 5.42× in efficiency.
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.