StarGAN v2 阅读笔记

@article{RN12,
author = {Choi, Yunjey and Uh, Youngjung and Yoo, Jaejun and Ha, Jung-Woo},
title = {StarGAN v2: Diverse Image Synthesis for Multiple Domains},
journal = {arXiv preprint arXiv:1912.01865},
year = {2019},
type = {Journal Article}
}

Contribution

StarGAN v2 在 StarGAN 的基础上进行了改进，解决了由一个域图像转换到目标域的多种图像，并支持多个目标域的问题。

Important Points

1. 设计了Mapping Network用于生成风格编码，摆脱了标签的束缚；
2. 用风格编码器指导Mapping Network进行目标风格学习，可以实现目标域下多风格图像的转换；
3. 公开了动物面部数据集AFQH，实现了图像翻译下较好的结果。

Motivation

Existing methods have limited diversity or multiple models for all domains, which are the issues that StarGAN v2 trys to address.