A Survey on Generative Adversarial Networks: Variants, Applications, and Training

The Generative Models have gained considerable attention in the field of\nunsupervised learning via a new and practical framework called Generative\nAdversarial Networks (GAN) due to its outstanding data generation capability.\nMany models of GAN have proposed, and several practical applications emerged in\nvarious domains of computer vision and machine learning. Despite GAN's\nexcellent success, there are still obstacles to stable training. The problems\nare due to Nash-equilibrium, internal covariate shift, mode collapse, vanishing\ngradient, and lack of proper evaluation metrics. Therefore, stable training is\na crucial issue in different applications for the success of GAN. Herein, we\nsurvey several training solutions proposed by different researchers to\nstabilize GAN training. We survey, (I) the original GAN model and its modified\nclassical versions, (II) detail analysis of various GAN applications in\ndifferent domains, (III) detail study about the various GAN training obstacles\nas well as training solutions. Finally, we discuss several new issues as well\nas research outlines to the topic.\n