Institutional Repository of Vilnius University
Neuroniniais tinklais paremtas vaizdų stiliaus perkėlimas realiu laiku žaidimuose /
Abstract
dc:descriptionExperiments using convolutional neural networks are carried out and new applications of style transfer in real-time game development are presented based on results from the experiments. This paper raises problems of applying style transfer real-time in video games and makes recommendations for solving the resulting problems. The experiments lead to new areas of application, such as the use of style transfer in prototyping and in the creation of unique animated textures. Using style transfer models, game developers can change the theme of an image or an object with a couple of clicks to help imagine the final look of the game environment or object. The experiment also identified a problem with style transfer described by other notable authors, which is that different frames of the style-transferred images from real-time view are temporally inconsistent. The Recommendations section provides solutions to this problem and the third experiment describes a way to exploit this problem by using it to create unique animated textures. This way of creating animated textures does not require writing additional code as conventional methods do. The results show comparisons between two different convolutional neural networks for style transfer. This paper highlights the advantages and disadvantages of these convolutional neural networks for style transfer and highlights which tasks these networks are more effective at.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Masolas, Augustas,
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:ELABAETD146241776&prefLang=en_US