I had applied to Bertelsmann Technology Scholarships and was among the Top 1,600 performers. As a result of which I was awarded a scholarship to Deep Learning Nanodegree program. A module of my program was dedicated to fake face generation using general adversarial networks (GANs). Figure above displays the generated samples (or generated faces) from my developed GAN model.
The generated samples do not exactly resemble human faces, but rather seem to be scary caricatures. This was because the training data consists of most images that are clipped below the chin area and were biased towards white celebrity faces. To improve the model work needs to be done towards creating a dataset comprising of high resolution images of complete faces of diverse ethnicity.