With Keras, I built a generator and a discriminator. The generator starts generating human faces from scratch. The discriminator has the input of pre-processed real human faces and will return the loss of the fake pictures generated by the generator. In the next round of iteration, the output of the discriminator is taken as the input of the generator. In the last batch of the last epoch, both my generator and my discriminator had a loss of around 1. Generated human faces is plotted in the Jupyter Notebook in the unit of one batch.
yiransii / generate-human-faces Goto Github PK
View Code? Open in Web Editor NEWGenerative Adversarial Networks: human faces generation