Evolution of Images with Diversity and Constraints Using a Generative Adversarial Network.

Generative Adversarial Networks (GANs) are a machine learning approach that have the ability to generate novel images. Recent developments in deep learning have enabled a generation of compelling images using generative networks that encode images with lower-dimensional latent spaces. Nature-inspired optimisation methods has been used to generate new images. In this paper, we train GAN with aim of generating images that are created based on optimisation of feature scores in one or two dimensions. We use search in the latent space to generate images scoring high or low values feature measures and compare different feature measures. Our approach successfully generate image variations with two datasets, faces and butterflies. The work gives insights on how feature measures promote diversity of images and how the different measures interact.