Improving Sample Quality of Diffusion Models Using Self-Attention Guidance
This work addresses the challenge of generating higher-quality images in diffusion models for AI and computer vision applications, representing an incremental improvement over existing guidance methods.
The paper tackles the problem of improving image quality in diffusion models by introducing Self-Attention Guidance (SAG), a condition- and training-free method that uses self-attention maps to adversarially blur attended regions, resulting in enhanced performance across models like ADM, IDDPM, Stable Diffusion, and DiT, with further gains when combined with conventional guidance.
Denoising diffusion models (DDMs) have attracted attention for their exceptional generation quality and diversity. This success is largely attributed to the use of class- or text-conditional diffusion guidance methods, such as classifier and classifier-free guidance. In this paper, we present a more comprehensive perspective that goes beyond the traditional guidance methods. From this generalized perspective, we introduce novel condition- and training-free strategies to enhance the quality of generated images. As a simple solution, blur guidance improves the suitability of intermediate samples for their fine-scale information and structures, enabling diffusion models to generate higher quality samples with a moderate guidance scale. Improving upon this, Self-Attention Guidance (SAG) uses the intermediate self-attention maps of diffusion models to enhance their stability and efficacy. Specifically, SAG adversarially blurs only the regions that diffusion models attend to at each iteration and guides them accordingly. Our experimental results show that our SAG improves the performance of various diffusion models, including ADM, IDDPM, Stable Diffusion, and DiT. Moreover, combining SAG with conventional guidance methods leads to further improvement.