Institutional Repository of Vilnius University
Vaizdo žaidimo eismo vaizdų generavimas naudojant nesuporuotų vaizdų transformacijas /
Abstract
dc:descriptionA method of generating video game traffic images is presented, using a pair of generative adverserial networks and an image segmentation model, thus creating a method capable of transforming the image to the desired domain while preserving the content of domain's the most important class. In this paper, it is described how using this method is it possible to transform images from a real traffic domain into a video game traffic domain while separately transforming its most important image class - cars. Additionally, results are compared against recently released pretrained diffusion models. The created method, while capable of generating better images than a standalone GAN under certain conditions, is less stable and is worse by evaluation metrics. While diffusion models generate images with much less noise and preserve details much better, they tend to either change the content too much or barely change it at all.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2023
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Oksas, Gytis,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD192847139&prefLang=en_US