guided diffusion

Guided diffusion is a deep learning algorithm that is used for image denoising and image inpainting tasks. It is based on the concept of diffusion processes, where the image is modeled as a diffusion of information through a set of random walks. The algorithm is guided by a conditioning image, which provides additional information to the diffusion process. This conditioning image can be any image that is related to the input image, such as a corrupted version of the input image or a semantic segmentation map.

Guided diffusion is a generative model, which means that it can be used to generate new images. By randomly sampling from the diffusion process, the algorithm can generate new images that are similar to the input image but have different details or textures. This makes it a useful tool for image synthesis tasks, such as generating realistic images of objects or scenes.

Guided diffusion has been shown to achieve state-of-the-art performance on several image denoising and inpainting benchmarks, and it has been applied to a wide range of image processing tasks, including super-resolution, colorization, and style transfer.