prikaz prve stranice dokumenta Super rezolucija temeljena na generativnim suparničkim neuronskim mrežama
  Download
PDF 2.84 MB
master's thesis
Super rezolucija temeljena na generativnim suparničkim neuronskim mrežama
Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2024. urn:nbn:hr:200:445057
Momić, Ines
Josip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
Department of Computer Engineering and Automation
Chair of Automation and Robotics

Institutional repository: Faculty of Electrical Engineering, Computer Science and Information Technology Osijek

Cite this document

Momić, I. (2024). Super rezolucija temeljena na generativnim suparničkim neuronskim mrežama (Master's thesis). Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek. Retrieved from https://urn.nsk.hr/urn:nbn:hr:200:445057

Momić, Ines. "Super rezolucija temeljena na generativnim suparničkim neuronskim mrežama." Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2024. https://urn.nsk.hr/urn:nbn:hr:200:445057

Momić, Ines. "Super rezolucija temeljena na generativnim suparničkim neuronskim mrežama." Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2024. https://urn.nsk.hr/urn:nbn:hr:200:445057

Momić, I. (2024). 'Super rezolucija temeljena na generativnim suparničkim neuronskim mrežama', Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, accessed 25 December 2024, https://urn.nsk.hr/urn:nbn:hr:200:445057

Momić I. Super rezolucija temeljena na generativnim suparničkim neuronskim mrežama [Master's thesis]. Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek; 2024 [cited 2024 December 25] Available at: https://urn.nsk.hr/urn:nbn:hr:200:445057

I. Momić, "Super rezolucija temeljena na generativnim suparničkim neuronskim mrežama", Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, Osijek, 2024. Available at: https://urn.nsk.hr/urn:nbn:hr:200:445057

Metadata
TitleSuper rezolucija temeljena na generativnim suparničkim neuronskim mrežama
Title (english)Super resolution based on generative adversarial neural networks
AuthorInes Momić
MentorRatko Grbić (mentor)
Committee memberRatko Grbić (predsjednik povjerenstva)
Committee memberMario Vranješ (član povjerenstva)
Committee memberMarijan Herceg (član povjerenstva)
GranterJosip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
(Department of Computer Engineering and Automation)
(Chair of Automation and Robotics)
Osijek
Defense date and country2024-09-30, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Artificial Intelligence
Abstract
Diplomski rad opisuje algoritme za super rezoluciju slika korištenjem generativnih suparničkih mreža. U radu su detaljno testirani i analizirani postojeći algoritmi za super rezoluciju te je razvijeno vlastito rješenje za super rezoluciju lica. Za izradu rješenja korištene su tehnike strojnog učenja, poput generativnih suparničkih mreža koje se sastoje od mreže generatora i mreže diskriminatora. Mreže se treniraju istovremeno te neprestano uče jedna od druge. Mreža generatora nastoji generirati realistične visoko rezolucijske izlaze iz nisko rezolucijskih ulaza, dok mreža diskriminatora procjenjuje jesu li generirane slike dovoljno realistične. Razvijena mreža trenirana je na FFHQ skupu podataka, a razvijeni model testiran je na CelebAHQ skupu podataka. Nakon evaluacije rezultata na metrikama za procjenu kvalitete slike, model je implementiran na ugradbeni računalni sustav NVIDIA Jetson Nano gdje je testiran za rad u stvarnom vremenu. Pomoću NVIDIA Jetson Nano platforme omogućen je prijenos uživo sa kamere te je razvijeni model uspješno pokazao mogućnost stvaranja visoko rezolucijskih slika lica u stvarnom vremenu. Daljnji napredak ovog algoritma može se postići treniranjem neuronske mreže na višestrukom broju grafičkih procesora, što bi osiguralo veću dostupnu memoriju i poboljšalo ukupnu izvedbu modela.
Abstract (english)
The thesis describes algorithms for super resolution using generative adversial networks (SRGAN). In this paper, the existing algorithms for super resolution were tested and analyzed in detal and a new solution for face super resolution was developed. Machine learning techniques, such as GANs were used to create the solution. GANs consist of two neural networks : generator network and discriminator network. Both networks are trained simultaneously and constatntly learn from each other. The generator aims to generate realistic high-resolution outputs from low-resolution inputs, while the discriminator evaluates whether the generated images are realistic enough. The developed network was trained on the FFHQ dataset, while the developed model was tested on the CelebAHQ dataset. After evaluating the results using image quality assessment metrics (IQA), the model was implemented on a NVIDIA Jetson Nano embedded system where it was tested in real-time. Using the NVIDIA Jetson Nano platform, live stream from the camera was enabled, and the developed model successfully demonstrated the possibility of creating high-resolution facial images in real time. Further progress of this algorithm can be achieved by training the neural network on a multiple number of GPUs, which would ensure more available memory and improve the overall performance of the model.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:200:445057
Study programmeTitle: University Study Automobile Computer and Communication
Study programme type: university
Study level: graduate
Academic / professional title: sveučilišni/a magistar/magistra inženjer/inženjerka automobilskog računarstva i komunikacija (sveučilišni/a magistar/magistra inženjer/inženjerka automobilskog računarstva i komunikacija)
Type of resourceText
File originBorn digital
Access conditionsOpen access
Terms of useLicence In copyright icon
Created on2024-10-04 09:15:02