A really interesting paper, and some good results. I was wondering if diffusion models could be trained with different types of noise - motion blur, geometric distortion, etc not just random latent perturbations. Conceptually it is quite a different problem to denoising!
A really interesting paper, and some good results. I was wondering if diffusion models could be trained with different types of noise - motion blur, geometric distortion, etc not just random latent perturbations. Conceptually it is quite a different problem to denoising!