Primjena maskirajućih autoenkodera za semantičku segmentaciju

master's thesis

• Title (english): Application of masked autoencoders for semantic segmentation
• Author: Luka Merćep
• Mentor: Siniša Šegvić (mentor)
• Committee member: Siniša Šegvić (predsjednik povjerenstva)
• Committee member: Marko Čupić (član povjerenstva)
• Committee member: Tomislav Hrkać (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2023-07-11, Croatia

Abstract

Detekcija anomalija područje je istraživanja sa primjenama u brojnim područjima. Modeli bazirani na transformerima trenutno su stanje tehnike u području obrade prirodnog jezika. Ovaj rad u detekciji anomalija koristi model maskirnog autoenkodera koji je baziran na arhitekturi vizualnog transformera. Vizualni transformer je model s minimalnim izmjenama nad arhitekturom modela za primjene transformera na slikama. U ovom radu korišten je skup podataka, za detekciju anomalija u industrijskim procesima, MVTec AD za učenje i evaluaciju modela. Cilj rada bio je ispitati efikasnost modela baziranog na arhitekturi maskirnog autoenkodera u području detekcije anomalija.

Abstract (english)

Anomaly detection is a field of research with applications in numerous areas. Models based on transformers are the current state of the art in the field of natural language processing. This work in anomaly detection uses the masked autoencoder model, which is based on the visual transformer architecture. The visual transformer is a model with minimal changes to the model architecture for transformer applications in images. In this work, a set of data, for the detection of anomalies in industrial processes, MVTec AD was used for model learning and evaluation. The aim of the work was to examine the effectiveness of the model based on the masked autoencoder architecture in the area of anomaly detection.

• Keywords: duboko učenje; detekcija anomalija; vizualni transformer; maskirni autoenkoder; industrijski procesi
• Keywords (english): deep learning; anomaly detection; visual transformer; masked autoencoder; industrial processes
• Language: croatian
• URN:NBN: urn:nbn:hr:168:211716
• 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: 2023-10-11 09:48:12