Institutional Repository of Vilnius University
Giliųjų klastočių atpažinimo išvengimas /
Abstract
dc:descriptionIn this work I present how deepfakes are created, how they are detected and also evaded. Deepfake creation research papers always excretes five common categories of deepfakes: face syntesis, face attribute manipulation, identity swap and also face expression swap. These categories are detailed in later sections. Also this work explains how the deepfakes are detected and in research papers common ways are spacial detection, detection based on image frequency, detection based on biological signals. In this research paper I dive deep into how deepfakes detection are evaded. First, I create the dataset for testing purpose then I find the deepfake detectors that I can use to test, prepare my attack against deepfake detectors and analize the results. The models that I found were to be working were „Xception “, „Resnext “, „DFDC 3D & 2D“and „MesoNet“. These models were tested and from the results it is clear that deepfake detectors cannot be used in real life situations because they are vulnerable to simple attacks like fast gradient sign method, blur attacks and compression attacks.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2021
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Vabolis, Ainis,
- Contributors dc:contributor
-
- Valaitis, Vytautas
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:ELABAETD107115506&prefLang=en_US