Institutional Repository of Vilnius University
Giliųjų klastočių atpažinimo metodai naudojant vaizdinius artefaktus /
Abstract
dc:descriptionDeepfake detection methods are deep learning models that execute binary classification tasks to expose visual artifacts. In order to determine the effectiveness of deepfake detection model’s tests are performed on well-known deepfake datasets, but these tests by itself are not sufficient enough to find out the model’s real-world use cases. In this paper an architectural analysis was conducted on deepfake detection methods. In order to measure EfficientNet’s and MesoNet’s recognition efficiency tests were conducted using a new data set, consisting of various long and short-term deepfake videos that were found online. Both models showed better performance results with shorter videos, however when tested with black-and-white data samples EfficientNet’s model showed remarkable improvements without increasing classification errors.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2022
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Petkevičius, Vincentas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD146241900&prefLang=en_US