Self-Modulation: ON SELF MODULATION FOR GENERATIVE ADVERSARIAL NETWORKS

Background

  1. 外部的条件(side information:class, label)有利于提高GAN的表现
  2. 但是这些条件往往不存在

 

Motivation

  1. 用无监督的方式取代外部条件

 

Model

Self-Modulation: ON SELF MODULATION FOR GENERATIVE ADVERSARIAL NETWORKS

将BN自带的参数β和γ用随机向量z控制的参数β‘和γ‘

β‘和γ‘由以z为输入的MLP得到(使用更复杂的架构不会更好)

Self-Modulation: ON SELF MODULATION FOR GENERATIVE ADVERSARIAL NETWORKS--->Self-Modulation: ON SELF MODULATION FOR GENERATIVE ADVERSARIAL NETWORKS

同样可以和外部条件y同时使用

Self-Modulation: ON SELF MODULATION FOR GENERATIVE ADVERSARIAL NETWORKS

Self-Modulation: ON SELF MODULATION FOR GENERATIVE ADVERSARIAL NETWORKS

上一篇:json中load和loads区别


下一篇:css3---经典代码