Denim: Heterogeneous Compute-in-Memory Accelerator Exploiting Denoising–Similarity for Diffusion Models https://sscs.ieee.org/wp-content/themes/movedo/images/empty/thumbnail.jpg 150 150 https://secure.gravatar.com/avatar/8fcdccb598784519a6037b6f80b02dee03caa773fc8d223c13bfce179d70f915?s=96&d=mm&r=g
Abstract:
Diffusion models have recently revolutionized the field of image synthesis due to their ability to generate photorealistic images. However, one of the main drawbacks of diffusion models is that the image generation process is expensive. Large image-to-image networks have to be applied multiple times in order to iteratively optimize the …