Neuronski prijenos stila

master's thesis

• Title (english): Neural style transfer
• Author: Rea Tot
• Mentor: Ivan Dunđer (mentor)
• Committee member: Sanja Seljan (predsjednik povjerenstva)
• Committee member: Vjera Lopina (član povjerenstva)
• Committee member: Ivan Dunđer (član povjerenstva)
• Granter: University of Zagreb; Faculty of Humanities and Social Sciences; (Department of information and Communication sciences); Zagreb
• Defense date and country: 2024-06-27, Croatia
• Scientific / art field, discipline and subdiscipline: SOCIAL SCIENCES; Information and Communication Sciences

Abstract

Umjetna inteligencija je posljednjih godina napredovala značajnom brzinom i tako postala jedna od važnijih tema u suvremenom svijetu. Jedna grana umjetne inteligencije istaknula se kao vrlo značajnom i utjecajnom – duboko učenje. Duboko učenje je moćan alat unutar umjetne inteligencije koji je omogućio stvaranje umjetnih neuronskih mreža, a s njima i konvolucijskih neuronskih mreža. Konvolucijske neuronske mreže vrlo su učinkovite u analizi i prepoznavanju vizualnih uzoraka na slikama, što se može vidjeti u procesu neuronskog prijenosa stila. Neuronski prijenos stila je tehnika koja kombinira stil jedne slike sa sadržajem druge što rezultira stiliziranom i vizualno privlačnom slikom. Cilj ovog diplomskog rada je bio pružiti teorijski okvir i cjelovito razumijevanje koncepta neuronskog prijenosa stila. Ovaj diplomski rad sastoji se od dva dijela: teorijskog i praktičnog. Teorijski dio pruža pregled umjetne inteligencije i strojnog učenja, kao i pregled koncepta dubokog učenja. Također, dubinski se ulazi u teorijski i praktični aspekt neuronskog prijenosa stila. Praktični dio diplomskog rada odnosi se na implementaciju neuronskog prijenosa stila uz pomoć prethodno treniranog modela. Implementacija je provedena korištenjem biblioteke za strojno učenje TensorFlow u okruženju Google Colaboratory. Rezultati implementacije prikazani su u zadnjem dijelu rada.

Abstract (english)

In recent years, artificial intelligence has rapidly advanced and has become one of the most important topics in the modern world. One branch of artificial intelligence has stood out as really influential – deep learning. Deep learning is a powerful tool within artificial intelligence that has allowed for the creation of artificial neural networks, and with them, convolutional neural networks. Convolutional neural networks are highly effective in analyzing and recognizing visual patterns in images, which can be seen in neural style transfer. Neural style transfer is a technique that combines the style of one image with the content of another, resulting in a stylized and visually appealing image. The aim of this thesis was to provide a theoretical framework and a comprehensive understanding of the concept of neural style transfer. This thesis is comprised of two parts: a theoretical and a practical part. The theoretical part provides an overview of artificial intelligence and machine learning, as well as an overview of the concept of deep learning. Also, it delves into the theoretical and practical aspects of neural style transfer. The practical part of the thesis is concerned with the implementation of neural style transfer using a pre-trained model. This implementation is done using the TensorFlow machine learning library in the Google Colaboratory environment. The results of the implementation are presented in the last part of the thesis.

• Keywords: umjetna inteligencija; duboko učenje; konvolucijske neuronske mreže; neuronski prijenos stila; Python
• Keywords (english): artificial intelligence; deep learning; convolutional neural networks; neural style transfer; Python
• Language: croatian
• URN:NBN: urn:nbn:hr:131:301795
• Study programme: Title: Information Sciences; specializations in: Archivistics Study, Library Science, Information Sciences (research), Information Sciences (teaching), Museology and heritage management, Informatology; Course: Information Sciences (research); Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra informacijskih znanosti (magistar/magistra informacijskih znanosti)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Open access
• Created on: 2024-06-27 13:48:34