GAN系列：Image-to-Image Translation with Conditional Adversarial Networks

Abstract 1. image-to-image trainslation 存在的兩個問題： 1）many-to-one:將照片映射爲edges/segments/semantic labels 2）one-to-many:將labels/sparse user inputs映射爲真實圖片 一般的做法是predict pixels form pixels；作者的創新點爲develop a co

>>阅读原文<<