Project Description:
We have access to a prior distribution, from which we can sample images. Given incomplete or degraded measurements sampled from this prior distribution, our objective is to reconstruct the missing or corrupted parts of the data. This problem applies to two specific tasks:
Image Inpainting: Given an image with masked regions, we aim to generate the missing portions using a diffusion-based posterior sampling approach.
Image Deblurring: Given blurred images sampled from the prior distribution, our goal is to restore sharp images by reversing the degradation process.