基于CycleGAN的图像风格转换方法及应用

CycleGAN based image style transfer method and its application

CycleGAN算法能很好地解决在图像未配对的情况下不能完成图像风格转换的问题,并且CycleGAN通过加入循环一致性损失,解决了数据分布捕捉能力不够的问题。实现了基于CycleGAN的图像风格转换算法的Web服务,并基于该Web服务开发了图像风格转换App。结果表明,使用CycleGAN算法可以较好地实现图像风格转换。

The CycleGAN algorithm can well solve the problem that the image style transfer cannot be completed when the images are not paired. In addition, CycleGAN solves the problem of the insufficient data distribution capture ability by adding cycle consistency loss. In this paper, a Web service with CycleGAN's image style transfer algorithm is realized, and an image style transfer App based on the Web service is developed thereby. Results show that CycleGAN algorithm can be used to achieve better image style conversion.