Semantička segmentacija slika korištenjem dubokih modela

master's thesis

• Title (english): Semantic image segmentation using deep models
• Author: Karla Ćosić-Dragan
• Mentor: Zoran Kalafatić (mentor)
• Committee member: Zoran Kalafatić (predsjednik povjerenstva)
• Committee member: Tomislav Hrkać (član povjerenstva)
• Committee member: Vladimir Čeperić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-02-26, Croatia

Abstract

Semantička segmentacija u kontekstu računalnog vida provodi dodjeljivanje klase objekta svakom pikselu slike u ovisnosti kojem objektu taj piksel pripada. Za taj se zadatak većinom koriste koder-dekoder arhitekture koje kao okosnicu obično koriste neke od standardnih arhitektura dubokog učenja. U ovom je radu predstavljen pregled nekih od standardnih arhitektura dubokog učenja, zatim pregled koder-dekoder arhitektura i metrika korištenih za semantičku segmentaciju. Za eksperimente je odabrana SwiftNet arhitektura uz programski okvir Pytorch Lightning i skupovi podataka Cityscapes i ACDC. Postignuti su očekivani rezultati i provedene su predikcije na validacijskim skupovima

Abstract (english)

Semantic segmentation in the context of computer vision assigns an object class to each pixel on the image depending on which object that pixel belongs to. The encoder-decoder architectures are mostly used for this task along with some of the standard deep learning architectures used as a backbone. This master thesis presents an overview of some standard deep architectures, followed by an overview of encoder-decoder architectures and metrics used for semantic segmentation. SwiftNet architecture with Pytorch Lightning framework were used for the experiments along with Cityscapes and ACDC datasets. Expected results were achieved and predictions were made on validation sets.

• Keywords: duboko učenje; semantička segmentacija; koder-dekoder arhitekture; Pytorch Lightning; skup podataka Cityscapes
• Keywords (english): deep learning; semantic segmentation; encoder-decoder architectures; Pytorch Lightning; Cityscapes dataset
• Language: croatian
• URN:NBN: urn:nbn:hr:168:882144
• 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: 2024-03-28 07:45:06