作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2019, Vol. 45 ›› Issue (9): 222-234. doi: 10.19678/j.issn.1000-3428.0051964

• 图形图像处理 • 上一篇    下一篇

生成式对抗网络研究与应用进展

柴梦婷, 朱远平   

  1. 天津师范大学 计算机与信息工程学院, 天津 300387
  • 收稿日期:2018-06-29 修回日期:2018-09-04 出版日期:2019-09-15 发布日期:2019-09-03
  • 作者简介:柴梦婷(1992-),女,硕士研究生,主研方向为图像处理、模式识别;朱远平,教授、博士。
  • 基金资助:
    国家自然科学基金(61602345,61703306);天津市科技计划项目(14RCGFGX00847)。

Research and Application Progress of Generative Adversarial Networks

CHAI Mengting, ZHU Yuanping   

  1. College of Computer and Information Engineering, Tianjin Normal University, Tianjin 300387, China
  • Received:2018-06-29 Revised:2018-09-04 Online:2019-09-15 Published:2019-09-03

摘要: 基于零和博弈思想的生成式对抗网络(GAN)可通过无监督学习获得数据的分布,并生成较逼真的数据。基于GAN的基础概念及理论框架,研究各类GAN模型及其在特定领域的应用情况,从数据相似性度量、模型框架、训练方法3个方面进行分析,对GAN改进与扩展的相关研究成果进行总结,并从图像合成、风格迁移等应用领域展开讨论,归纳出GAN的优势与不足,同时对其应用前景进行展望。分析结果表明,GAN的学习能力与可塑性强,改进潜力大,应用范围广,但其发展面临的挑战是训练过程不稳定,且缺乏生成数据质量的客观评价标准。

关键词: 生成式对抗网络, 生成式模型, 对抗学习, 深度学习, 人工智能

Abstract: The Generative Adversarial Networks(GAN) based on the zero-sum game idea can obtain the distribution of data through unsupervised learning and generate more realistic data.Based on the basic concepts and theoretical framework of the generated confrontation network,the GAN models and the application results in specific fields are studied,and the data similarity measure,model framework and training method are summarized.The research and related research results of the improvement and expansion are analyzed,and the practical application fields such as image synthesis and style migration are discussed.The advantages and disadvantages of GAN are summarized,and the application prospects are prospected.Analysis results show that the GAN has strong learning ability and plasticity,great potential for improvement and wide application range.However,its development challenges are unstable training process and lack of objective evaluation criteria for generating data quality.

Key words: Generative Adversarial Networks(GAN), generative model, adversarial learning, deep learning, artificial intelligence

中图分类号: