In the previous section, we looked at generating digits using a vanilla generator and a discriminator network. However, we can have a scenario where the network can learn the features in an image much better by using the convolution architectures, as the filters in a CNN learn specific details within an image. Deep Convolutional Generative Adversarial Networks (DCGANs) take advantage of this phenomenon to come up with new images.
Generating images using a Deep Convolutional GAN
How to do it...
While the intuition of how a DCGAN works is very similar to that of a GAN (which we worked with in the previous recipe), the major difference is in the architecture of the generator and discriminator of the DCGAN, which looks as follows...
