mix-precision quantization

Denim: Heterogeneous Compute-in-Memory Accelerator Exploiting Denoising–Similarity for Diffusion Models

Denim: Heterogeneous Compute-in-Memory Accelerator Exploiting Denoising–Similarity for Diffusion Models 150 150

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 …

View on IEEE Xplore