Samonadzirano učenje stereoskopske rekonstrukcije pseudooznačavanjem

master's thesis

• Title (english): Self-supervised learning of stereoscopic reconstruction through pseudo-labeling
• Author: David Kerman
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Alan Jović (član povjerenstva)
• Committee member: Tomislav Hrkać (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-11, Croatia

Abstract

Ovaj rad uvodi samonadzirani pristup učenju stereoskopske rekonstrukcije pseudooznačavanjem, temeljen na prethodnim radovima. Pružamo pregled temeljnih principa
stereoskopske rekonstrukcije i relevantnih modela dubokog učenja. Rad uključuje detaljan opis samonadziranog učenja pseudooznačavanjem, zajedno s našim inovativnim
tehnikama filtriranja osmišljenim za poboljšanje točnosti pseudooznaka. Nadalje, demonstriramo poboljšanje performansi nadziranog stereoskopskog modela postignuto korištenjem rafiniranih pseudooznaka. Eksperimentalni rezultati su predstavljeni kako bi
se potvrdila učinkovitost našeg pristupa. Identificirani su problemi i predložene modifikacije koje bi poboljšale trenutnu metodu.

Abstract (english)

This work introduces a self-supervised approach to stereoscopic reconstruction through
pseudo-labeling, based on the previous studies. We provide an overview of the fundamental principles of stereoscopic reconstruction and relevant deep learning models.
The methodology includes a detailed description of self-supervised learning via pseudo-
labeling, along with our innovative filtering techniques designed to enhance the accuracy of pseudo-labels. We demonstrate the performance improvements in the supervised
stereo model achieved using these refined pseudo-labels. Experimental results are presented to validate the effectiveness of our approach. Finally, we identified problems with
the current method and proposed changes which could potentially improve model accuracy.

• Keywords: stereoskopska rekonstrukcija; duboko učenje; samonadzirano učenje; računalni vid; pseudooznačavanje
• Keywords (english): stereoscopic reconstruction; deep learning; self-supervised learning; com- puter vision; pseudo-labeling
• Language: english
• URN:NBN: urn:nbn:hr:168:969377
• 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: 2024-10-12 21:03:14