prikaz prve stranice dokumenta Segmentacija objekata
No public access
master's thesis
Segmentacija objekata
Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing, 2019. urn:nbn:hr:168:466123
Pranjić, Marko
University of Zagreb
Faculty of Electrical Engineering and Computing

Institutional repository: FER Repository

Cite this document

Pranjić, M. (2019). Segmentacija objekata (Master's thesis). Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing. Retrieved from https://urn.nsk.hr/urn:nbn:hr:168:466123

Pranjić, Marko. "Segmentacija objekata." Master's thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, 2019. https://urn.nsk.hr/urn:nbn:hr:168:466123

Pranjić, Marko. "Segmentacija objekata." Master's thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, 2019. https://urn.nsk.hr/urn:nbn:hr:168:466123

Pranjić, M. (2019). 'Segmentacija objekata', Master's thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, accessed 27 July 2024, https://urn.nsk.hr/urn:nbn:hr:168:466123

Pranjić M. Segmentacija objekata [Master's thesis]. Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing; 2019 [cited 2024 July 27] Available at: https://urn.nsk.hr/urn:nbn:hr:168:466123

M. Pranjić, "Segmentacija objekata", Master's thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, 2019. Available at: https://urn.nsk.hr/urn:nbn:hr:168:466123

Metadata
TitleSegmentacija objekata
Title (english)Object Segmentation
AuthorMarko Pranjić
MentorTomislav Pribanić (mentor)
Committee memberTomislav Pribanić (predsjednik povjerenstva)
Committee memberMarko Subašić (član povjerenstva)
Committee memberHrvoje Džapo (član povjerenstva)
GranterUniversity of Zagreb
Faculty of Electrical Engineering and Computing
Zagreb
Defense date and country2019-07-16, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Abstract
U radu se primjenjuje metoda dubokog učenja za problem segmentacije ljudskog tijela. Nakon teorijskog uvoda u segmentaciju objekata opisana je metoda rješavanja problema. Korišten je skup podataka Freiburg Sitting People Dataset i arhitektura mreže temeljena na TernausNet-16 modelu te kombinacija unakrsne entropije i Dice koeficijenta za funkciju gubitka. Poslije opisa navedene metode prikazani su rezultati provedenog eksperimenta u kojem je dosegnut mIOU od 51.15%, ali i otkriven problem prenaučenosti na boju obuće. Takod̄er su razmotreni eksperimenti koji su prethodili odabranoj metodi koji navode na korištenje predtrenirane mreže za koder segmentacijske mreže. Kao zaključak predlaže se korištenje regularizacije i korištenje većeg skupa podataka uz bolju raspodjelu primjera po skupovima za učenje, validaciju i evaluaciju.
Abstract (english)
In this thesis deep learning is used for solving the problem of human body part segmentation. After a theoretical introduction to object segmentation, a description of the used method is given. The chosen dataset which is used is Freiburg Sitting People Dataset, the network architecture is based on the TernausNet-16 model, and the loss function is a combination of categorical cross-entropy and Dice coefficient. The next section shows the result of the experiment in which the mIOU score of 51.15% was achieved, and the problem of overfitting on shoe color became apparent. Experiments that led to choosing a pre-trained network for the encoder of the segmentation network are also discussed. As a conclusion, it is recommended to use regularization and a larger dataset along with a better split of data among train, validation and evaluation set.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:168:466123
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 on2020-01-07 21:32:48