DCGAN is initialized with random weights, so a random code plugged in the network would create a totally random image. However, while you may think, the network has a lot of parameters that we will tweak, and the purpose is to find a location of those parameters that makes samples created from random codes seem like the training knowledge.8MB of SR