Generating Gorillas.
This is an implementation of a Deep Convolutional Generative Adversarial Network. It has two primary parts: a generator and a discriminator.
The generator attempts to create images like the training dataset (pictures of gorrilas), while the discriminator assesses whether the generated images are distinguishable from the training dataset.
These two parts work together to produce increasingly accurate generated images.
Because there are no labels, I assessed outputs subjectively.
Time to generate gorillas:
25 epochs:
Later, I tried to implement a Super-Resolution Generative Adversarial Network, which functions to improve the resolution of images. In this case, our low resolution gorilla pictures.
Here were the best three looking examples (after many iterations):
