Semantička segmentacija primjeraka korištenjem dubokih konvolucijskih modela

master's thesis

• Title (english): Semantic Segmentation of Instances Using Deep Convolutional Models
• Author: Matej Balun
• Mentor: Tomislav Hrkać (mentor)
• Committee member: Tomislav Hrkać (predsjednik povjerenstva)
• Committee member: Zoran Kalafatić (član povjerenstva)
• Committee member: Željka Mihajlović (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-02-19, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Segmentacija instanci problem je suvremenog računalnog vida i specifično domene dubokog učenja. Sama problematika kombinacije je klasifikacije, semantičke segmentacije, detekcije i lokalizacije objekata, dajući tako kompletnu informaciju o klasama objekata i pojedinim instancama tih klasa, u obliku okvira, koordinata, segmentnih maski i labela klasa i njihovih instanci. Facebook AI research team (FAIR) u članku iz 2017. godine predlaže model za segmentaciju instanci pod imenom Mask R-CNN. Model je baziran na prijašnjim modelima za zadaću detekcije objekata poput R-CNN, Fast R-CNN, Faster R-CNN itd. U ovome radu proučavana je implementacija i performanse modela na skupu uličnih urbanih prizora pod imenom Cityscapes, u kojem model daje daleko bolje rezultate nego ostali pristupi u problemu segmentacije instanci na tom skupu, ali i otvara mjesta za daljnja poboljšanja i nove te učinkovitije pristupe.

Abstract (english)

Instance segmentation is a problem of modern computer vision and specifically the domain of deep learning. The problem is a combination of classification, semantic segmentation, detection, and localization of objects, thus giving complete information about object classes and individual instances of those classes, in the form of frames, coordinates, segment masks, and labels of classes and their instances. Facebook AI research team (FAIR) in a 2017 article proposes a model for instance segmentation called Mask R-CNN. The model is based on previous models for object detection tasks such as R-CNN, Fast R-CNN, Faster R-CNN, etc. This paper studies the implementation and performance of the model on a set of street urban scenes called Cityscapes, in which the model gives far better results than other approaches to the problem of instance segmentation in that set, but also opens room for further improvements and new and more efficient approaches.

• Keywords: Računalni vid; strojno učenje; duboko učenje; segmentacija instanci; semantička segmentacija; detekcija objekata; lokalizacija objekata; klasifikacija obje- kata; Mask R-CNN; COCO; Cityscapes; konvolucijske neuronske mreže
• Keywords (english): Computer vision; machine learning; deep learning; instance segmentation; semantic segmentation; object detection; object localization; object classification; Mask R-CNN; COCO; Cityscapes; convolutional neural networks
• Language: croatian
• URN:NBN: urn:nbn:hr:168:575012
• 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 09:43:21