prikaz prve stranice dokumenta 3D segmentacija objekata
No public access
master's thesis
3D segmentacija objekata
Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing, 2020. urn:nbn:hr:168:790226
Jertec, Andrej
University of Zagreb
Faculty of Electrical Engineering and Computing

Institutional repository: FER Repository

Cite this document

Jertec, A. (2020). 3D segmentacija objekata (Master's thesis). Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing. Retrieved from https://urn.nsk.hr/urn:nbn:hr:168:790226

Jertec, Andrej. "3D segmentacija objekata." Master's thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, 2020. https://urn.nsk.hr/urn:nbn:hr:168:790226

Jertec, Andrej. "3D segmentacija objekata." Master's thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, 2020. https://urn.nsk.hr/urn:nbn:hr:168:790226

Jertec, A. (2020). '3D segmentacija objekata', Master's thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, accessed 27 December 2024, https://urn.nsk.hr/urn:nbn:hr:168:790226

Jertec A. 3D segmentacija objekata [Master's thesis]. Zagreb: University of Zagreb, Faculty of Electrical Engineering and Computing; 2020 [cited 2024 December 27] Available at: https://urn.nsk.hr/urn:nbn:hr:168:790226

A. Jertec, "3D segmentacija objekata", Master's thesis, University of Zagreb, Faculty of Electrical Engineering and Computing, Zagreb, 2020. Available at: https://urn.nsk.hr/urn:nbn:hr:168:790226

Metadata
Title3D segmentacija objekata
Title (english)3D Object Segmentation
AuthorAndrej Jertec
MentorTomislav Pribanić (mentor)
Committee memberTomislav Pribanić (predsjednik povjerenstva)
Committee memberTomislav Petković (član povjerenstva)
Committee memberHrvoje Džapo (član povjerenstva)
GranterUniversity of Zagreb
Faculty of Electrical Engineering and Computing
Zagreb
Defense date and country2020-07-10, Croatia
Scientific / art field,
discipline and subdiscipline		TECHNICAL SCIENCES
Computing
Abstract
Područje 3D dubokog učenja je brzo razvijajuće područje sa sve većim brojem znanstvenih radova i istraživanja. Karakterizira ga manjak podataka za treniranje, te je u tom smjeru u ovom radu predstavljen postupak za efikasno geneniranje podataka segmentiranog ljudskog tijela pomoću SMPL modela uz minimalnu potrebu za ručnim označavanjem. Nadalje, generirani sintetički podaci su korišteni za treniranje PointNet neuronske mreže na zadatku segmentacije ljudskog tijela. Validacija neuronske mreže je obavljena na ukupno 3 skupa podataka od kojih su 2 sintetička te jedan realni skup podataka skeniranih ljudskih tijela. U radu je pokazano da je korištenjem samo sintetičkih podataka uz adekvatno uzorkovanje i normalizaciju moguće dobiti robusnu neuronsku mrežu koja dobro generalizira na realnom skupu podataka.
Abstract (english)
The area of 3D deep learning is expanding field of study with growing number of scientific papers and research projects. Currently it is being limited by the lack of training datasets. To remedy this problem, in this work we present simple and efficient procedure of using SMPL model for generation of dataset of segmented human body with minimal needs for manual segmentation. Furthermore, we use the synthetic dataset for training PointNet neural network in solving the task of human body segmentation. We validate the neural network on 3 different datasets. Two of which are synthetic and one that contains scans of real human body. In this work we show that using only synthetic dataset with proper sampling and normalization one can train robust neural network that performs and generalizes well on real data.
Keywords
Keywords (english)
Languagecroatian
URN:NBNurn:nbn:hr:168:790226
Study programmeTitle: Information and Communication Technology
Study programme type: university
Study level: graduate
Academic / professional title: magistar/magistra inženjer/inženjerka informacijske i komunikacijske tehnologije (magistar/magistra inženjer/inženjerka informacijske i komunikacijske tehnologije)
Type of resourceText
File originBorn digital
Access conditionsClosed access
Terms of useLicence In copyright icon
Public note
Created on2023-01-13 09:10:40