Frontalizacija lica temeljena na dubokim generativnim modelima

master's thesis

• Title (english): Face Frontalization Using Deep Generative Models
• Author: Patrik Mesec
• Mentor: Alan Jović (mentor)
• Committee member: Alan Jović (predsjednik povjerenstva)
• Committee member: Ante Đerek (član povjerenstva)
• Committee member: Marin Golub (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2022-07-06, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Suvremeni modeli identifikacije lica imaju manju točnost na profilnim pozama lica, posebice u ekstremnim slučajevima. Glavni cilj rada bio je izrada modela za ostvarivanje prednjeg pogleda na lice (frontalizacija) s ciljem poboljšanja točnosti identifikacije lica. Najvažniji dio rada bio je sakupljanje kritične količine podataka pomoću kojih je učeno nekoliko dubokih generativnih modela. Najbolji naučeni model temelji se na arhitekturi modela FFWM. Učene su različite varijacije modela FFWM, te... More je prikazan utjecaj pojedinih komponenti modela. Točnost identifikacije lica mjerena je na više različitih modela za izlučivanje značajki lica na skupu za evaluaciju. Frontalizacijom lica ostvarena su značajna poboljšanja identifikacije lica na svim modelima. Osim kvantitativne evaluacije, dana je i vizualizacija ostvarenih rezultata frontalizacije na skupu za evaluaciju i na dodatnim primjercima iz različitih izvora. Less

Abstract (english)

Modern face identification models have worse accuracy on the profile poses of the face, especially in extreme cases. The main goal of this thesis was to develop a model for the face frontalization with the aim of improving the accuracy of face identification. The most important part of the work was the collection of a critical amount of data by which several deep generative models were trained. The best trained model is based on the architecture of the FFWM model. Different variations of the... More FFWM model were trained and the influence of some model components was shown. The accuracy of face identification was measured on several different models for extracting facial features on validation set. Face frontalization led to a significant improvement of face identification on all models. In addition to quantitative evaluation, the face frontalization results achieved on the validation set and additional examples from various sources were also visualized. Less

• Keywords: identifikacija lica; frontalizacija lica; duboki generativni modeli; model FFWM; podaci za učenje
• Keywords (english): face identification; face frontalization; deep generative models; FFWM model; training data
• Language: croatian
• URN:NBN: urn:nbn:hr:168:880501
• 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: 2023-01-14 09:32:06