Semantička segmentacija s brzom pažnjom u preskočnim vezama

master's thesis

• Title (english): Semantic segmentation with fast attention in skip connections
• Author: Lovro Katalinić
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Marko Čupić (član povjerenstva)
• Committee member: Vladimir Čeperić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-07-16, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Semantička segmentacija je zadatak računalnog vida koji elemente slike razvrstava u različite semantičke razrede. U posljednje vrijeme kvalitetna rješenja zadatka daju konvolucijski modeli koji se sastoje od osnove za klasifikaciju i ljevičastog naduzorkovanja. Modeli za obradu slike u stvarnom vremenu moraju žrtvovati svoju kompleksnost kako bi dobili na brzini. Mehanizam brze pažnje prilagođen je za obradu u stvarnom vremenu i poboljšava preciznost modela. U radu su u preskočne veze modela SwiftNet uvedeni moduli brze pažnje. Dobiveni model evauliran je nad skupom podataka Cityscapes i uspoređen s osnovnim modelom.

Abstract (english)

Semantic segmentation is a computer vision task which classifies pixels of an image into different semantic classes. Recently convolutional models based on classification backbone and encoder-decoder architecture have given quality solutions for the task. Models for real-time inference have to sacrifice their complexity in order to increase their speed. Fast attention mechanism is adapted for real-time inference and it improves model accuracy. In this thesis, fast attention modules have been introduced into lateral connections of SwiftNet model. The resulting model was evaluated over the Cityscapes dataset and compared
with the baseline model.

• Keywords: semantička segmentacija; zaključivanje u stvarnom vremenu; brza pažnja; koder-dekoder; SwiftNet; FANet; preskočne veze
• Keywords (english): semantic segmentation; real-time inference; fast attention; encoder-decoder; SwiftNet; FANet; skip connections
• Language: croatian
• URN:NBN: urn:nbn:hr:168:117491
• 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 12:38:56