Semantička segmentacija osoba dubokim konvolucijskim modelom

master's thesis

• Title (english): Semantic Segmentation of Humans Using a Deep Convolutional Model
• Author: Elena Vukelić
• Mentor: Marin Šilić (mentor)
• Committee member: Marin Šilić (predsjednik povjerenstva)
• Committee member: Klemo Vladimir (član povjerenstva)
• Committee member: Goran Delač (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2022-07-01, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Zadatak semantičke segmentacije podjela je objekta na prepoznatljive dijelove (segmente) kako bi se mogli lakše analizirati. U radu je objašnjen temelj razvoja takvih modela te naveden predstavnik koji je osnovni gradivni blok mnogih, a to je konvolucijska neuronska mreža. Nakon toga ulazi se u područje segmentacije, gdje su opisani neki postojeći modeli i njihova poboljšanja. Proveden je eksperiment koji koristi unaprijed trenirani Mask R-CNN model iz Detectron2 platforme. Skup podataka za učenje modela odabran je iz osobne kolekcije autorice, a proces pripreme i označavanja detaljno je opisan. Prikazan je i proces evaluacije, odnosno implementacija prilagođenog IoU evaluatora te dobivena rješenja i analiza rezultata.

Abstract (english)

The task of Semantic Segmentation is to divide an object into recognisable parts (segments) so that they can be analysed more easily. This thesis explains the basis for developing such models and lists the representative as the basic building block of many: the Convolutional Neural Network. Afterwards, one enters the segmentation area, where some existing models and their improvements are described. An experiment using a pre-trained Mask R-CNN model from the Detectron2 platform was conducted. The data set for learning the model was selected from the author's collection, and the process of preparation and labelling is described in detail. The evaluation process is also presented, i.e. the implementation of a customised IoU evaluator, and the obtained solutions and analysis of the results are described.

• Keywords: Duboko učenje; konvolucijska neuronska mreža; semantička segmentacija; segmentacija osoba; Detectron2; Mask R-CNN; IoU
• Keywords (english): Deep Learning; Convolutional Neural Network; Semantic Segmentation; Segmentation of Humans; Detectron2; Mask R-CNN; IoU
• Language: croatian
• URN:NBN: urn:nbn:hr:168:469427
• 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: 2023-01-13 14:15:20