DCGAN is initialized with random weights, so a random code plugged in the network would crank out a totally random image. Nonetheless, when you might imagine, the network has countless parameters that we can tweak, along with the objective is to locate a setting of those parameters that makes samples produced from random codes appear to be the inst