Prenos slikarskega stila s pomočjo globokih nevronskih mrež

RAKITA, ALJOŠA (author), Solina, Franc (mentor), Batagelj, Borut (comentor)

V tej diplomski nalogi, je najprej razloženo kaj so sploh globoke nevronske mreže in kratka razlaga njihovega delovanja. Nato je nekaj besed nemenjenih temu kako je stil slike definiran in kaj vse predstavlja. Po tem, je natančno razloženo in analizirano delovanje specifične metode za prenos stila izbrane slike na ciljno sliko s pomočjo globokih nevronskih mrež. Temu sledi demonstracija delovanja, kjer smo naredili "lažne" slike s pomočjo stila znanih slovenskih impresionistov. Zaključili pa smo s spletno anketo, ki je preverila kako realne so producirane slike, to je ali so naključno izbrani ljudje sposobni ločiti tako generirane "lažne" slike od pravih slik slovenskih impresionistov.

• Language: Slovenian
• Keywords: prenos stila, globoke nevronske mreže, analiza
• Work type: Bachelor thesis/paper
• Organization: FRI - Faculty of Computer and Information Science
• Year: 2018
• PID: 20.500.12556/RUL-103536
• Publication date in RUL: 19.09.2018

RAKITA, ALJOŠA, 2018, Prenos slikarskega stila s pomočjo globokih nevronskih mrež [online]. Bachelor’s thesis. [Accessed 5 April 2025]. Retrieved from: https://repozitorij.uni-lj.si/IzpisGradiva.php?lang=eng&id=103536

Secondary language

• Language: English
• Title: Painting style transfer with deep neural networks

Abstract

What are deep neural networks and how they work is explained first in this diploma thesis. Next, we describe what is a painting style and what determines the style. With deep neural networks one can transfer the painting style from a selected source image to another target image. We demonstrate style transfer by making ``fake’’ pictures using paintings of Slovene impressionist painters. At the end we made a web survey to find out, if randomly selected people could distinguish such ``fake’’ pictures from the real painting of slovenian impressionists.

• Keywords: style transfer, deep neural network, analysis