This repository contains the official implementation of the DDRM paper from NeurIPS 2022. It adapts pre-trained DDPMs (Denoising Diffusion Probabilistic Models) to solve general linear inverse problems in image restoration without requiring task-specific supervised training. The method uses the generative prior of diffusion models to perform efficient restoration across different tasks including deblurring, inpainting, and super-resolution, working directly with off-the-shelf pretrained models from OpenAI and other sources.