Chapter 5 - A review of the techniques of images using GAN

Abstract

The generative adversarial networks (GANs) have attracted substantial awareness as they have the potential to generate the bulk of the unlabeled data. With the limited availability of the training data in the supervised learning process for the deep learning methods, it becomes complicated to train the model effectively. This limitation of the training dataset increases the need for unsupervised feature learning techniques. It does not require the annotation as well as the labeling of the data. So the GAN techniques for the domain adaptation, image-to-image translation, data augmentation, and image generation are a boon to the applications that require a large amount of the data for research purposes. Over the years, along with the original GAN, the different GANs such as deep convolution GANs (DCGANs), conditional GANs (CGANs), StackGAN, Wasserstein GANs (WGANs), Cyclical GANs, etc. have been derived. This chapter presents a review of the techniques for generation of images using GAN.