Back to search

Institutional Repository of Vilnius University

Objektų defektų aptikimas vaizduose naudojant mokymąsi iš keleto pavyzdžių /

Abstract

dc:description

Defect detection in objects, which plays a crucial role in industrial quality control, has recently shifted from manual inspection to deep learning-based methods. However, the application of these methods is often hindered by the limited training data. The aim of this work is to develop and evaluate an ensemble of anomaly detection methods based on few-shot learning, and evaluate it on MVTec LOCO AD dataset which is designed for detecting structural and logical anomalies. The proposed ensemble combines improved versions of the WinCLIP and MAEDAY metods: WinCLIP localizes potential anomalies, while MAEDAY uses this information to generate a reconstruction error map. The outputs of both methods are then combined into a final anomaly score. Experimental results show that the ensemble achieved higher anomaly detection accuracy than each of individual methods. The best results are achieved in detecting structural anomalies when using four training images. Compared to other anomaly detection methods that rely on learning from all available data, the proposed ensemble demonstrates higher accuracy but lower efficiency in the context of few-shot learning.

Degree

thesis:*
Grantor dc:publisher
Institutional Repository of Vilnius University
Year dc:date
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Kondratjeva, Aleksandra,

Rights

dc:rights
Statement dc:rights
  • info:eu-repo/semantics/openAccess
Language dc:language
lit

Identifiers

dc:identifier.*

Chain of custody

source
Harvested from
Vilnius University
Base URL
epublications.vu.lt/oai
Last updated
2026-11-19
Source record
OAI-PMH GetRecord
related terms
citation

Kondratjeva, Aleksandra,. Objektų defektų aptikimas vaizduose naudojant mokymąsi iš keleto pavyzdžių /. Institutional Repository of Vilnius University, 2025. https://repository.vu.lt/VU:ELABAETD239031249&prefLang=en_US