Slabo nadzirana semantička segmentacija primjeraka primjenom suparničkog učenja

master's thesis

• Title (english): Weakly supervised instance segmentation by adversarial learning
• Author: Katarina Blažić
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (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: 2019-07-15, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Semantička segmentacija prirodnih scena je važan zadatak računalnog vida. Skup postupak pribavljanja oznaka na razini piksela potrebnih za nadzirani pristup semantičkoj segmentaciji je potaknuo razvoj slabo nadziranih pristupa. U ovom radu su opisani neki od slabo nadziranih pristupa za semantičku segmentaciju koji se baziraju na dubokom učenju kao i primjene suparničkog učenja za semantičku segmentaciju. U sklopu rada je implementiran model suparničkog učenja za slabo naziranu semantičku segmentaciju primjeraka zasnovan na pravokutnim okvirima. Generator koristi značajke naučenog Faster R-CNN detektora uz pomoć kojih gradi segmentacijske maske, a diskriminator pruža signale za učenje. Za učenje i validaciju modela su korišteni automobili iz Cityscapes skupa podataka te su prikazani dobiveni rezultati.

Abstract (english)

Semantic segmentation of natural scenes is an important task of computer vision. Due to expensive pixel-wise annotations required for fully supervised semantic segmentation various weakly supervised approaches have been proposed. In this thesis, several weakly supervised semantic segmentation approaches based on deep learning have been described as well as applications of adversarial learning to semantic segmentation. Adversarial learning model for object instance segmentation based on bounding boxes is implemented. Generator uses learned Faster R-CNN features and constructs a segmentation mask while discriminator provides learning signals. Model was trained and evaluated on cars form Cityscapes dataset and the results are shown.

• Keywords: suparničko učenje; semantička segmentacija; segmentacija primjeraka; slabo nadzirano učenje; duboko učenje
• Keywords (english): adversarial learning; semantic segmentation; instance segmentation; weakly-supervised learning; deep learning
• Language: croatian
• URN:NBN: urn:nbn:hr:168:259143
• 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 16:50:41