Authors : Mustafa Radif
Abstract: The study presents a new training method for Generative Adversarial Networks (GANs). The new method is about adding low-resolution layers to images in the networks and gradually increasing the resolution until high resolution with good and realistic images are developed. The method helps to create stable images of good quality and the speed of training also, increases by a factor of 2. A score of 8.8 was achieved for unsupervised CIFAR 10. The study presents details of the implementation where rivalry and competition between the two networks are reduced.
Mustafa Radif , 2019. Using Generative Adversarial Networks (GANs) to Generate Facial Attributes. Journal of Engineering and Applied Sciences, 14: 9073-9085.