prikaz prve stranice dokumenta Optimizacija segmentacije cestovnih scena u stvarnom vremenu
  Download
PDF 1.43 MB
master's thesis
Optimizacija segmentacije cestovnih scena u stvarnom vremenu
Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, 2024. urn:nbn:hr:200:757367
Barišić, Renata
Josip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
Department of Software Engineering
Chair of Visual Computing

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

Cite this document

Barišić, R. (2024). Optimizacija segmentacije cestovnih scena u stvarnom vremenu (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:200:757367

Barišić, Renata. "Optimizacija segmentacije cestovnih scena u stvarnom vremenu." 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:200:757367

Barišić, Renata. "Optimizacija segmentacije cestovnih scena u stvarnom vremenu." 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:200:757367

Barišić, R. (2024). 'Optimizacija segmentacije cestovnih scena u stvarnom vremenu', Master's thesis, Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek, accessed 27 November 2024, https://urn.nsk.hr/urn:nbn:hr:200:757367

Barišić R. Optimizacija segmentacije cestovnih scena u stvarnom vremenu [Master's thesis]. Osijek: Josip Juraj Strossmayer University of Osijek, Faculty of Electrical Engineering, Computer Science and Information Technology Osijek; 2024 [cited 2024 November 27] Available at: https://urn.nsk.hr/urn:nbn:hr:200:757367

R. Barišić, "Optimizacija segmentacije cestovnih scena u stvarnom vremenu", 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:200:757367

Metadata
TitleOptimizacija segmentacije cestovnih scena u stvarnom vremenu
Title (english)Optimization of segmentation of road scenes in real time
AuthorRenata Barišić
MentorČaslav Livada (mentor)
Committee memberČaslav Livada (predsjednik povjerenstva)
Committee memberAlfonzo Baumgartner (član povjerenstva)
Committee memberTomislav Galba (član povjerenstva)
GranterJosip Juraj Strossmayer University of Osijek
Faculty of Electrical Engineering, Computer Science and Information Technology Osijek
(Department of Software Engineering)
(Chair of Visual Computing)
Osijek
Defense date and country2024-09-26, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Artificial Intelligence
Abstract
Cilj ovog diplomskog rada bio je implementirati U-Net, SegNet i ENet arhitekture te njihov trening uz pomoć PyTorch biblioteke, a dobivene rezultate koristiti za usporedbu navedenih arhitektura po pitanju točnosti, brzini izvođenja, učinkovitosti u stvarnom vremenu i kompleksnosti implementacije. Trening je proveden na Cityscapes skupu podataka, na kojem je provedena pred-obrada potrebna za korištenje pri treningu. Prije treninga, arhitekture su uspoređene po veličini i broju parametara; tijekom treninga, prate se srednja vrijednost unije i presjeka te točnost piksela; a nakon treninga uspoređuju se po trajanju i brzini treniranja, broju epoha te se provodi testiranje na testnom dijelu skupa podataka. Pri testiranju se mjere vrijeme zaključivanja i mogućnost izvođenja u stvarnom vremenu, a točnost se mjeri kvalitativno, pošto ne postoji izvor istine za segmentacije testnog skupa. Navedenim usporedbama pružen je dubok uvid u primjenjivost navedenih arhitektura u kontekstu automobilske percepcije te je svakoj arhitekturi pronađena prikladna primjena na temelju njezinih rezultata i performansi – ENet je izabrana za zadatke u kojima su ključne brzina i učinkovitost resursa, a U-Net i SegNet se, zbog svojih sličnih rezultata, predlažu za istu primjenu, a to su situacije u kojima je točnost bitnija od iskorištenosti resursa.
Abstract (english)
The aim of this thesis was to implement U-Net, SegNet and ENet architectures along with their training with the help of the PyTorch library, and using the obtained results to compare the mentioned architectures in terms of accuracy, execution speed, real-time efficiency and implementation complexity. The training was carried out on the Cityscapes dataset, which was pre-processed for use in training. Before the training, the architectures were compared according to their size and number of parameters; during the training the mean Intersection over Union and the pixel accuracy were monitored; after the training, they were compared according to the duration and speed of the training, the number of epochs, and testing was carried out on the test part of the data set. During testing, inference time and real-time-execution capabilities were measured quantitatively, and accuracy was measured qualitatively, since there is no ground truth for the test set segmentations. The aforementioned comparisons provided deep insight into the applicability of the architectures in the context of automotive perception, and a suitable task was found for each architecture based on its results and performance - ENet was chosen for tasks where speed and resource efficiency are key, while U-Net and SegNet are suggested for the same tasks due to their similar results – tasks in which the accuracy is more important than resource utilization.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:200:757367
Study programmeTitle: Graduate University Study Programme in Computer Engineering
Study programme type: university
Study level: graduate
Academic / professional title: sveučilišni/a magistar/magistra inženjer/inženjerka računarstva (sveučilišni/a magistar/magistra inženjer/inženjerka računarstva)
Type of resourceText
File originBorn digital
Access conditionsOpen access
Terms of useLicence In copyright icon
Created on2024-10-04 07:56:36