Image generation by generative adversarial networks

U ovom radu proučava se primjena generativnih suparničkih mreža (GAN) za generiranje novih slika bespilotnih letjelica. U uvodnom dijelu rada daje se pregled osnovnih koncepata GAN-a, teorije igara i procesa treniranja te opisi različitih arhitektura i varijanti GAN-a. Metodologija rada uključuje odabir odgovarajućeg algoritma za generiranje slika baziranog na GAN-ovima, s posebnim naglaskom na DCGAN i WGAN te njihove prednosti za generiranje slika bespilotnih letjelica. Također se opisuju skup podataka, eksperimentalno okruženje i implementacija. Rezultati dobiveni kroz implementaciju i treniranje GAN-a su prikazani i analizirani. Kroz zaključak se sumiraju rezultati i daje se pregled ostvarenih ciljeva u radu.In this paper, we research the application of generative adversarial networks (GANs) for generating new images of unmanned aerial vehicles (UAVs). The introductory section provides an overview of the basic concepts of GANs, game theory, and the training process, along with descriptions of different architectures and variants of GANs. The methodology of the paper includes selecting an appropriate GAN-based algorithm for image generation, with a specific focus on DCGAN and WGAN, highlighting their advantages for generating UAV images. Additionally, the dataset, experimental environment, and implementation are described. The results obtained through the implementation and training of the GAN are presented and analyzed. The conclusion summarizes the findings and provides an overview of the achieved objectives in the paper