Gradient Domain Diffusion Models for Image Synthesis

TL;DR

This paper introduces a novel approach to image synthesis using diffusion models in the gradient domain, which accelerates convergence due to mathematical equivalence and sparsity, enhancing efficiency across various applications.

Contribution

The paper proposes performing diffusion in the gradient domain, leading to faster convergence and broader applicability in image processing and related fields.

Findings

Gradient domain diffusion models converge faster than traditional models.

Numerical experiments confirm increased efficiency of the proposed method.

Applicable to diverse image processing and computer vision tasks.

Abstract

Diffusion models are getting popular in generative image and video synthesis. However, due to the diffusion process, they require a large number of steps to converge. To tackle this issue, in this paper, we propose to perform the diffusion process in the gradient domain, where the convergence becomes faster. There are two reasons. First, thanks to the Poisson equation, the gradient domain is mathematically equivalent to the original image domain. Therefore, each diffusion step in the image domain has a unique corresponding gradient domain representation. Second, the gradient domain is much sparser than the image domain. As a result, gradient domain diffusion models converge faster. Several numerical experiments confirm that the gradient domain diffusion models are more efficient than the original diffusion models. The proposed method can be applied in a wide range of applications such as image processing, computer vision and machine learning tasks.