GANS¶

简介 ¶

Generative adversarial networks

Catch Me If You Can

Acknowledgement¶

理解生成对抗网络 (GANs) - 知乎

一文看尽深度学习中的生成对抗（GAN）网络 _energy-basedgan-CSDN 博客

图解 生成对抗网络 GAN 原理 超详解 _gan 原理图 -CSDN 博客

算法原理 ¶

损失函数 ¶

训练过程 ¶

评价 ¶

优点 ¶

缺点 ¶

Variants¶

Wasserstein GANs¶

Critics: the discriminator is called a critic,it tells how real the data is

Conditional GANs | cGANs¶

cycle GANs¶

replace the main object but keep the main features

StyleGANs¶

seperate different style of the images

Challenges¶

• Vanishing gradients: the generator gradients vanish, making it impossible to train the generator
• Mode collapse: the generator collapses to a single point in the data space( 一张图片效果好就一直生成 )
• Convergence

代码实战 ¶