ThinkNotClear
  • Home
  • Categories
  • Tags
  • Archives

StyleGAN: A Style-Based Generator Architecture for Generative Adversarial Networks

目录

Table of Contents
  • 概述
目录

概述

  1. bilinear up/down sampling, 更长的训练, 调参
  2. 应用 AdaIN 到 Progressive GAN
  3. w = 8层MLP(z), 再用 A(w)->scale,bias of AdaIN
  4. 移除过去的 z输入给conv, 变成 固定的
  5. Noise 作为 stochastic variantion 控制细节变化
  6. mixing regularization. 部分数据使用 style mixing生成的图片

stylegan


Published

Sep 18, 2022

Category

paper

Tags

  • GAN 3
  • generator 8
  • paper 9
  • Powered by Pelican. Theme: Elegant by Talha Mansoor