Lightweight diffusion models: a survey

DOI: https://doi.org/10.1007/s10462-024-10800-8

IF: 9.588

2024-05-31

Artificial Intelligence Review

Abstract:Diffusion models (DMs) are a type of potential generative models, which have achieved better effects in many fields than traditional methods. DMs consist of two main processes: one is the forward process of gradually adding noise to the original data until pure Gaussian noise; the other is the reverse process of gradually removing noise to generate samples conforming to the target distribution. DMs optimize the application results through the iterative noise processing process. However, this greatly increases the computational and storage costs in the training and inference stages, limiting the wide application of DMs. Therefore, how to effectively reduce the resource consumption of using DMs while giving full play to their good performance has become a valuable and necessary research problem. At present, some research has been devoted to lightweight DMs to solve this problem, but there has been no survey in this area. This paper focuses on lightweight DMs methods in the field of image processing, classifies them according to their processing ideas. Finally, the development prospect of future work is analyzed and discussed. It is hoped that this paper can provide other researchers with strategic ideas to reduce the resource consumption of DMs, thereby promoting the further development of this research direction and providing available models for wider applications.

computer science, artificial intelligence

What problem does this paper attempt to address?

### Problems the Paper Attempts to Solve

This paper aims to address the issue of high computational and storage costs associated with diffusion models (DMs) during both training and inference stages. Although diffusion models have achieved better results than traditional methods in many fields, their iterative noise processing significantly increases resource consumption, limiting their widespread application. Therefore, how to effectively reduce resource consumption while maintaining the good performance of diffusion models has become a valuable and necessary research topic.

Currently, some research has been dedicated to lightweight diffusion models to solve this problem, but there has not yet been a dedicated review article that systematically summarizes this field. This paper focuses on lightweight diffusion model methods in the field of image processing, classifying them according to different processing approaches and analyzing and discussing future development prospects. It is hoped that this paper can provide strategic ideas for other researchers to reduce the resource consumption of diffusion models, thereby promoting further development in this research direction and providing usable models for broader applications.