The discriminator's architecture is much more like what we've already seen in previous chapters. It's really just a typical image classifier, as shown in the following figure. The output is sigmoid because the discriminator will be predicting the probability that the input image is a member of the set of real images. The discriminator is solving a binary classification problem:

Now that we've covered the architecture of the DCGAN and it's various layers, lets take a look at how the framework is trained.