摘要
生成对抗网络(GAN)作为一种生成式模型,在计算机视觉领域每年有大量的论文产出。为帮助后续研究人员快速全面地了解GAN,并加以应用,本文首先从设计思想和网络架构两个方面介绍GAN的基本原理,再由其目标函数与训练过程分析GAN存在的缺陷及其原因;然后,根据改进角度的不同和应用两个方面对GAN的变体模型进行详述,总结了它们的改进之处、优点和局限性等;接着,分别从客观定量、主观定性和任务专项评估等角度对GAN生成图像的保真度和多样性进行归纳分析;最后,讨论了GAN衍生模型近年来的一些核心问题与最新研究进展,并分析了未来的发展趋势。
There are a significant number of research papers on Generative Adversarial Network(GAN),a generative model,each year in the field of computer vision.In order to help subsequent researchers to quickly and comprehensively understand and apply GAN,this paper firstly introduces the basic principles of GAN from the perspectives of design ideas and network architecture,and then analyzes the flaws of GAN and their reasons based on its objective functions and training processes.Secondly,it elaborates various derivative models of GAN from the perspectives of improvement and application,with their enhancements,advantages and limitations summarized.Thirdly,this paper summarizes and analyzes the quality and diversity of GAN generated images from the perspectives of subjective qualitative,objective quantitative and task-specific evaluation.Finally,some core issues and latest research progress of GAN series models in recent years are discussed and their future development trend is analyzed.
作者
肖锋
周雨洁
张文娟
XIAO Feng;ZHOU Yujie;ZHANG Wenjuan(School of Defence Science and Technology,Xi’an Technological University,Xi’an 710021,China;School of Computer Science and Engineering,Xi’an Technological University,Xi’an 710021,China;School of Sciences,Xi’an Technological University,Xi’an 710021,China)
出处
《西安工业大学学报》
2025年第4期652-673,共22页
Journal of Xi’an Technological University
基金
国家自然科学基金项目(62171361)。
关键词
计算机视觉
生成对抗网络
图像生成
图像转换
模式崩溃
computer vision
generative adversarial network
image generation
image translation
collapse mode
