Getting to know the generation confrontation network (11) -- using pytoch to build wgan to generate handwritten digits

〇、WGAN Introduce

About WGAN You can refer to the article ：GAN Introduction to generating countermeasure network and DCGAN、WGAN Wait for the introduction . Here's just to introduce WGAN With respect to code GAN or DCGAN The difference between .
In the paper 《Wasserstein GAN》 in , The author has deduced a lot of formulas and theorems , Finally, the improved algorithm implementation process is given , Compared with the original GAN However, only four changes have been made in the implementation process of the algorithm ：

• The last layer of the discriminator is removed sigmoid;
• Generator and discriminator loss No log;
• Every time the parameters of the discriminator are updated, their absolute values are truncated to no more than a fixed constant c;
• Don't use momentum based optimization algorithms （ Include momentum and Adam）, recommend RMSProp,SGD It's OK ;
  The algorithm is shown in the figure below ：
  

WGAN