The Generator architecture is a map from a low-dimensional space to a higher dimensional space G(z), wherein, fake images produced by the generator live. In other words, the generator samples z and projects it into some multidimensional, for example 
if the images produced by the Generator are 4 by 4. In fully convolutional Generator architectures, transposed convolutions are used in the Generator to increase the data-dimensionality. Our first Generator implementation will be based on DCGAN, a non-fully convolutional architecture. Let's start by analyzing the following diagram:
Generator DCGAN architecture. Source: Unsupervised Representation ...
