Sustav za semantičko unapređenje baza afektivne multimedije korištenjem modela dubokog učenja

master's thesis

• Title (english): System for the semantic enhancement of affective multimedia databases using deep learning models
• Author: Katarina Virkes
• Mentor: Marko Horvat (mentor)
• Committee member: Marko Horvat (predsjednik povjerenstva)
• Committee member: Gordan Gledec (član povjerenstva)
• Committee member: Krešimir Fertalj (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2024-07-09, Croatia

Abstract

Tema rada je izrada sustava za semantičko unapređenje baza afektivne multimedije korištenjem modela dubokog učenja. Baze afektivne multimedije su napravljene u srhu izazivanja emocija. Jedna od takvih baza je Nencki Affective Picture System, NAPS, koja sadrži 1356 slika raspoređeni u 5 kategorija. Baze uz multimediju imaju zapisanu semantiku te očekivane emocije koji se ručno zapisuju. Proučili smo neke od modela dubokog učenja za automatizirano zapisivanje semantike. Proučili smo kakve rezultate YOLO, You Only Look Once, algoritam daje u ovisnosti o verziji algoritma i setu na kojem je treniran. Vidjeli smo da odabrani set jako utječe na rezultate, dok novije verzije daju nešto bolje rezultate od starijih. Proučili smo i kakve rezultate daje gotov alat Azure AI Vision. Na kraju smo zaključili da bi ovakvi modeli mogli olakšat održavanje baza afektivne medije.

Abstract (english)

The topic of the paper is the creation of a system for semantic improvement of affective multimedia databases using deep learning models. Bases of affective multimedia are made with the aim of evoking emotions. One such database is the Nencki Affective Picture System, NAPS, which contains 1356 pictures arranged in 5 categories. Databases with multimedia have menually written semantics and expected emotions. We have explored some of the deep learning models for automated semantic writing. We studied what kind of results the YOLO, You Only Look Once, algorithm gives depending on the version of the algorithm and the set on which it was trained. We have seen that the selected set greatly affects the results, while the newer versions give slightly better results than the older ones. We also studied the results of the ready-made tool Azure AI Vision. In the end, we concluded that such models could facilitate the maintenance of databases of affective media.

• Keywords: afektivna multimedija; baze afektivne multimedije; računalni vid; detekcija objekata; duboko učenje; umjetni neuroni; umjetne neuronske mreže; konvolucijske neuronske mreže; YOLO
• Keywords (english): affective multimedia; affective multimedia databases; computer vision; object detection; deep learning; artificial neurons; artificial neural networks; convolutional neural networks; YOLO
• Language: croatian
• URN:NBN: urn:nbn:hr:168:907344
• 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: Open access
• Created on: 2024-10-12 22:25:02