《Progressive Growing of GANs for Improved Quality, Stability, and Variation》-论文阅读笔记
论文结构
1.Introduction
2.Progressive growing of GANs
3.Increasing variation using
minibatch standard deviation
4.Normalization in generator and
discriminator
4.1 Equalized learning rate
4.2 Pixelwise feature vector
normalization in generator
5.Multi-scale statistical similarity
for assessing GAN results
6.Experiments
6.1 Importance of individual contributions in
terms of statistical similarity
6.2 Convergence and training speed
6.3 High-resolution image generation using
CelebA-HQ dataset
6.4 LSUN results
6.5 CIFAR10 inception scores
7.Discussion
摘要
原文
We describe a new training methodology for generative adversarial networks. The key idea is to grow both the generator and discriminator progressively: starting from a low resolution, we add new layers that model increasingly fine details as training progresses. This both speeds the training up and greatly stabilizes it, allowing us to produce images of unprecedented quality, e.g., CelebA images at 1024^2. We also propose a simple way to increase the variation in generated images, and achieve a record inception score of 8.80 in unsupervised CIFAR10. Additionally, we describe several implementation details that are important for discouraging unhealthy competition between the generator and discriminator. Finally, we suggest a new metric for evaluating GAN results, both in terms of image quality and variation. As an additional contribution, we construct a higher-quality version of the CelebA dataset.
总结
1.使用渐进的方式来训练生成器和判别器：先从生成低分辨率图像开始，然后不断增加模型层数来提升生成图像的细节
2.这个方法能加速模型训练并大幅提升训练稳定性，生成前所未有的的高质量图像（1024*1024）
3.提出了一种简单的方法来增加生成图像的多样性
4.介绍了几种限制生成器和判别器之间不健康竞争的技巧
5.提出了一种评价GAN生成效果的新方法，包括对生成质量和多样性的衡量
6.构建了一个CELEBA数据集的高清版本
研究背景
Research background
生成式模型的类别
•显性密度模型
易解显性模型：定义一个方便计算的密度分布，主要的模型是Fully visible belief nets，简称FVBN，也被称作Auto-Regressive Network
近似显性模型：可以定义任意的密度分布，使用近似方法来求解
隐性密度模型
GAN
• 神经自回归网络（PixelRNN/CNN）
通过链式法则把联合概率分布分解为条件概率分布的乘积使用神经网络来参数化每个P
PixelRNN逐像素生成，效率很低，PixelCNN效果不如PixelRNN
• VAE-GAN
编码器：使P(z|x)逼近分布P(z)，比如标准正态分布，同时最小化生成器(解码器)和输入x的差距
解码器：最小化输出和输入x的差距，同时要骗过判别器
判别器：给真实样本高分，给重建样本和生成样本低分
图像生成的评价指标
• 可以评价生成样本的质量
• 可以评价生成样本的多样性，能发现过拟合、模式缺失、模式崩溃、直接记忆样本的问题
• 有界性，即输出的数值具有明确的上下界
• 给出的结果应当与人类感知一致
• 计算评价指标不应需要过多的样本
• 计算复杂度尽量低
