Deep convolutional GAN, also called DCGAN, is used to generate color images. Here we use a convolutional layer in the generator and discriminator. We'll also need to use batch normalization to get the GAN to train appropriately. We will discuss batch normalization in detail in the performance improvement of deep neural networks chapter. We'll be training GAN on the SVHN dataset; a small example is shown in the following figure. After training, the generator will be able to create images that are nearly identical to these images. You can download the code for this example: