Convolutional discriminator

Now we will see the architecture of a discriminator in DCGAN. As we know, the discriminator takes the image and it tells us whether the image is a real image or a fake image. Thus, it is basically a binary classifier. The discriminator is composed of a series of convolutional and batch norm layers with leaky ReLU activations.

The architecture of the discriminator is shown in the following diagram:

As you can see, it takes the input image of the 64 x 64 x 3 shape and performs a series of convolutional operations with a leaky ReLU activation function. We apply batch normalization at all layers except at the input layer.

Remember, we don't apply a max pooling operation in both the discriminator and the generator. Instead, we apply a strided convolution operation (that is convolution operation with strides).

In a nutshell, we enhance the vanilla GAN by replacing the feedforward network in the generator and the discriminator with the convolutional network.

..................Content has been hidden....................

You can't read the all page of ebook, please click here login for view all page.
Reset