Final project of the Deep Learning Nanodegree
Implemented a Deep Convolutional Generative Adversarial Neural Network to generate new human faces.The generator is fed with noise while the discriminator tries to classify real images and the generator output images as real or fake.This starts a competition between the discriminator and generator to outsmart each other.The concept has been taken from Game Theory.The generator consists of transpose convolutional layers while the discriminator is a regular convolutional network without max pool layers.
The Network was trained on MNIST and CelebA datasets.The epochs take a long time to train even on GPUs.However,the images generated by the network are realistic after 1 or 2 epochs.