Samonadzirane metode za vizualno raspoznavanje

master's thesis

• Title (english): Self-supervised methods for visual recognition
• Author: Vedran Bogdanović
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Jan Šnajder (član povjerenstva)
• Committee member: Marko Čupić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2022-09-19, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

U zadnje vrijeme, samonadzirane metode postižu rezultate koji su sumjerljivi s nadziranim metodama. Ove metode znaju davati i bolje rezultate od nadziranih u području vizualnog raspoznavanja jer se skaliraju s neoznačenim podacima, kojih je mnogo više. Ove metode su možda i bitnije u zadacima kod kojih je označavanje podataka jako skupo.
U ovom radu su predstavljene neke od trenutno popularnih samonadziranih metoda. Jedna od tih metoda je implementirana i trenirana na dostupnim skupovima podataka. Dobiveni rezultati su objašenje i uspoređeni s drugim popularnim metodama. Također je dotaknuta i mogučnost uporabe ove metode na zadatku guste predikcije.

Abstract (english)

Self-supervised methods have been producing results that are comparable to supervised methods. These methods even surpass many supervised methods in visual representation learning tasks because they scale with unlabeled data. These methods are also very important in tasks for which labeling data is very expensive, like dense prediction.
In this thesis some more popular contrastive self-supervised methods are described. One of those methods was implemented and trained. The results obtained are explained and compared to other state-of-the-art methods. The possibility of using that method for dense prediction was also touched.

• Keywords: Samonadzirano učenje; Kontrastivno učenje; Vizualno raspoznavanje; Gusta predikcija; Duboko učenje
• Keywords (english): Self-supervised learning; Contrastive learning; Visual recognition; Dense prediction; Deep learning
• Language: english
• URN:NBN: urn:nbn:hr:168:798722
• 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 14:35:29