Regional Style Transfer Based on Partial Convolution Generative Adversarial Network
Yi Min Yang, Cheng Wang, Lan Lin · 2020
Nowadays style transfer has become one of the major fields of image processing. Recently, many awesome methods have been proposed. Traditional research of style transfer forces on transformation of whole painting. However, research of more flexible regional style transfer, which transforms different parts of a picture into different styles, remain open. In this paper, we present a novel Generative Adversarial Network (GAN) based regional style transfer method, by which the style features of two style datasets could be automatically learned. By given a mask, our method can transfer part of the input content image to one style and rest to the other. Specifically, we preprocess two style datasets to extract style features by two training partial convolution-based discriminators. The experimental results show that our method achieves significant performance efficiently.