Uporaba konvolucijskih neuronskih mreža za kolorizaciju crno-bijelih slika

master's thesis

• Title (english): Using convolutional neural networks for colorizing black and white images
• Author: Marin Gregorić
• Mentor: Sanja Seljan (mentor)
• Committee member: Ivan Dunđer (predsjednik povjerenstva)
• Committee member: Vjera Lopina (član povjerenstva)
• Committee member: Sanja Seljan (član povjerenstva)
• Granter: University of Zagreb; Faculty of Humanities and Social Sciences; (Department of information and Communication sciences); Zagreb
• Defense date and country: 2023-09-18, Croatia
• Scientific / art field, discipline and subdiscipline: SOCIAL SCIENCES; Information and Communication Sciences; Information and Software Engineering

Abstract

Cilj ovog rada je izraditi sustav strojnog učenja za kolorizaciju crno-bijelih slika na osobnom računalu, osnovan na konvolucijskim neuronskim mrežama. U teorijskom dijelu rada prezentira se teorijska podloga neuronskih mreža s naglaskom na aktivacijske funkcije, funkcije gubitka i algoritme učenja neuronskih mreža. U praktičnom dijelu prikazana je izrada, treniranje i dizajn modela te sustava za kolorizaciju crno-bijelih slika. Također, izrađen je i poseban skup podataka za treniranje sustava za kolorizaciju. Detaljno je opisana arhitektura modela, te je model kontekstualiziran usporedbom sa drugim kolorizacijskim modelima. Dokazuje se da je moguće izraditi i trenirati sustav za kolorizaciju na osobnom računalu uz optimizaciju procesa treniranja. Također, dokazuje se da je moguće izraditi model koji je usporediv sa drugim modelima konvolucijskih neuronskih mreža koristeći nove metode i elemente arhitekture modela.

Abstract (english)

The goal of this paper is to create a machine learning system based on convolutional neural networks for colorization of black and white images on a personal computer. I the theoretical part of this paper neural networks are presented with an emphasis on activation functions, loss functions and learning algorithms of neural networks. The practical part of this paper describes building, designing and training of the model, and system for colorization of black and white images. Also, a novel dataset has been developed for training of colorization systems. Model architecture is detailed, and the model is contextualized by comparing it to other colorization models. This paper proves that it is possible to build and train a colorization system on a personal computer by optimizing the training process. Also, it is proven that it is possible to build a model that is comparable with other models based on convolutional neural networks by using new methods and elements of model architecture.

• Keywords: strojno učenje; neuronske mreže; konvolucijske neuronske mreže; aktivacijske funkcije; funkcije gubitka; metode učenja neuronskih mreža; kolorizacija
• Keywords (english): machine learning; neural networks; convolutional neural networks; activation functions; loss functions; neural network optimization algorithms; colorization
• Language: croatian
• URN:NBN: urn:nbn:hr:131:526919
• 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 (teaching); 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: 2023-09-19 17:24:15