Use of generative-adversarial networks when creating content

Abstract

The application of generative-adversarial networks in the creation of content is studied. Monitoring of training, analysis of architectures, determination of internal processes at the level of layers, research of properties of latent space, and interaction with it are carried out. Variants of using the specified networks in image generation are considered. Special attention is paid to practical implementation aspects, including selecting optimal parameters and data processing. The difference between a classifier and a discriminator is formulated. The principles of generative-adversarial networks and their influence on the efficiency and quality of generated images are studied. The advantages and limitations of using GANs in content creation are considered.