Generative Adversarial Networks Projects

Architectural details of a DCGAN

As mentioned before, the DCGAN network uses convolutional layers in both networks. To reiterate, a CNN is a network with convolution layers, followed by a normalization or pooling layers, followed by an activation function. In DCGANs, the discriminator network takes an image, downsamples it with the help of convolutional and pooling layers, and uses a dense classification layer to classify the image as real or fake. The generator network takes a random noise vector from latent space, upsamples it using upsampling mechanisms, and finally generates an image. We use Leaky ReLU as the activation function for the hidden layers and dropout between 0.4 and 0.7 to avoid overfitting.

Let's take a look at the configuration ...