Before moving forward, let's have a look at the architecture of the generator network:
Source: arXiv:1511.06434 [cs.LG]
The preceding diagram contains different layers in the architecture of the generator network and shows how it generates an image of a resolution of 64 x 64 x 3.
The DCGAN's generator network contains 10 layers. It performs strided convolution to increase the spatial resolution of the tensor. In Keras, a combination of upsampling and convolution layers is equal to a strided convolutional layer. Basically, the generator network takes a noise vector sampled from a uniform distribution and keeps ...
