The Generative Adverserial Networks were studied and implemented on generating faces using CelebA for training and then the second experiment was to generate realistic looking leaf images by being trained on a leaf dataset obtained from kaggle.
Here are the images that were saved after certain epochs. It shows gradual and impressive training.
Initially it showed clear noise and no training:
Then after it started to train and improve gradually as shown in the below images :
Then finally it started to show realistic looking images :
