Optimalne strukture podataka za 3D reprezentaciju objekata u visokoj rezoluciji

master's thesis

• Title (english): Optimal Data Structures for 3D Representations at High Resolution
• Author: Matej Gradiček
• Mentor: Tomislav Pribanić (mentor)
• Committee member: Tomislav Pribanić (predsjednik povjerenstva)
• Committee member: Marko Subašić (član povjerenstva)
• Committee member: Hrvoje Džapo (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2019-07-16, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Cilj ovog rada bio je proučiti način kako optimalno prezentirati 3D objekte i proučiti arhitekture nekih mreža za analizu prezentacije. Analizirana je stablasta struktura
octree, njezina implementacija i testiran je rad opisanih mreža kod klasifikacije modela iz baze podataka ModelNet10. Opisan je Keras, framework za duboko učenje i
pokušaj integracije octree strukture u slojeve Kerasa. Integracija nije uspjela jer Keras koristi tenzore kao strukturu nad kojom vrši operacije. Zbog toga je napisan mali
Python framework u kojemu se mreža testirala. Framework je napisan da bude sli-
čan Kerasu i da bude jednostavan za korištenje, nadogradiv i kako bi mogao koristiti
stablastu strukturu.

Abstract (english)

The goal of this thesis was to investigate optimal ways of representation of 3D
objects and investigate architectures of some deep learning networks for analyzing this
representation. Octree structure and its implementation were analyzed and networks
using octree structures were tested on task of classifying models from ModelNet10
database. Deep learning framework Keras was described and attempt of integrating
octree structure into Keras layers. Integration failed because Keras is using tensors as
data structures as structure on which operations are made. New Python framework for
deep learning was written and networks were tested on it. Framework was written to
be like Keras, easily extendable, simple to use and to be able to use octree structure.

• Keywords: Reprezentacija 3D objekata; duboko učenje; klasifikacija; octree struktura; Keras
• Keywords (english): Representation of 3D objects; deep learning; classification; octree structure; Keras
• Language: croatian
• URN:NBN: urn:nbn:hr:168:202761
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2020-01-07 18:11:50