| Abstract [eng] |
One of the most popular technologies today is artificial neural networks. This technology is widely used in medicine, policing and IT fields. The main idea behind neural networks is to train an algorithm on the basis of given data to perform certain tasks without programming specific actions, so that neural networks can be said to be able to come up with answers to certain given tasks. However, one of the unique human abilities that is more difficult to implement in computers is the ability to create something new and real. The human ability to imagine different worlds, environments, people, images, and to translate this into books, music, paintings, thus creating real new things. To replicate this ability using artificial intelligence in computers, various algorithms have been developed, including generative adversarial networks (GANs), which will be investigated in this thesis, as they are now one of the most popular algorithms used for photo generalization. In addition GANs can replicate the human ability to create new and realistic things such as photos of people, can merge two photos or generate a new unique photo, complete a part of a photo, generate a photo from text, increase the resolution of a photo, etc. This study will therefore look at where face generation is used, how the original GAN works, and what GANs are now developed to generate, merge or modify faces in videos. In addition, the performance of the GANs found will be reviewed and the results they generate will be compared between the intended categories. |
