prikaz prve stranice dokumenta Autoenkoderi za kompresiju slike
No public access
master's thesis
Autoenkoderi za kompresiju slike
Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing, 2024. urn:nbn:hr:168:107786
Kada, Karlo
University of Zagreb
Faculty of Electrical Engineering and Computing

Institutional repository: FER Repository

Cite this document

Kada, K. (2024). Autoenkoderi za kompresiju slike (Master's thesis). Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing. Retrieved from https://urn.nsk.hr/urn:nbn:hr:168:107786

Kada, Karlo. "Autoenkoderi za kompresiju slike." Master's thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, 2024. https://urn.nsk.hr/urn:nbn:hr:168:107786

Kada, Karlo. "Autoenkoderi za kompresiju slike." Master's thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, 2024. https://urn.nsk.hr/urn:nbn:hr:168:107786

Kada, K. (2024). 'Autoenkoderi za kompresiju slike', Master's thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, accessed 27 December 2024, https://urn.nsk.hr/urn:nbn:hr:168:107786

Kada K. Autoenkoderi za kompresiju slike [Master's thesis]. Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing; 2024 [cited 2024 December 27] Available at: https://urn.nsk.hr/urn:nbn:hr:168:107786

K. Kada, "Autoenkoderi za kompresiju slike", Master's thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, 2024. Available at: https://urn.nsk.hr/urn:nbn:hr:168:107786

Metadata
TitleAutoenkoderi za kompresiju slike
Title (english)Autoencoders for image compression
AuthorKarlo Kada
MentorTomislav Petković (mentor)
Committee memberTomislav Petković (predsjednik povjerenstva)
Committee memberMarko Subašić (član povjerenstva)
Committee memberJuraj Petrović (član povjerenstva)
GranterUniversity of Zagreb
Faculty of Electrical Engineering and Computing
Zagreb
Defense date and country2024-07-11, Croatia
Abstract
Ovaj rad istražuje upotrebu stanjivajućih autoenkodera za kompresiju slika ljudskog lica. Rad istražuje performanse autoenkodera s različitim brojevima filtera po sloju, fokusirajući se na trening autoenkodera korištenjem više-razinskog indeksa strukturalne sličnosti (MS-SSIM) umjesto tradicionalnih metoda poput srednje kvadratne pogreške (MSE). Korištenjem testnog podskupa uzetog iz skupa „Human Faces“ objavljenog na Kaggleu, evaluira se sposobnost autoenkodera da proizvede rekonstrukcije slika visoke kvalitete, s posebnim naglaskom na percipiranu kvalitetu slike. Rezultati pokazuju da je MS-SSIM adekvatna metrika za treniranje autoenkodera za kompresiju slika. Jedno od mogućih poboljšanja ovog modela je proširivanje funkcije gubitka za učenje modela mjerom krominancije kako bi se pospješila reprodukcija boje.
Abstract (english)
This thesis explores the use of slimmable compressive autoencoders for the compression of human face images. The thesis investigates the performance of autoencoders with varying numbers of filters per layer, focusing on training using the multi-scale structural similarity index measure (MS-SSIM) instead of traditional metrics like the mean squared error (MSE). By utilizing a subset of the “Human Faces” dataset published on Kaggle, the autoencoder's ability to produce high-quality image reconstructions is evaluated both quantitatively and qualitatively, with a particular emphasis on perceptual image quality. The findings demonstrate that MS-SSIM is an adequate metric for training autoencoders for image compression. Possible enhancement of the model could be the inclusion of a chrominance term in the training loss function to facilitate better color reproduction.
Keywords
Keywords (english)
Languageenglish
URN:NBNurn:nbn:hr:168:107786
Study programmeTitle: Computing
Study programme type: university
Study level: graduate
Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
Type of resourceText
File originBorn digital
Access conditionsClosed access
Terms of useLicence In copyright icon
Public note
Created on2024-10-12 21:00:53