Realtime Deepfake Detection using Video Vision Transformer

Abstract

Practically, Deepfake technology has given people access to generate fake videos that look like real content using neural networks, and can further create misconceptions and deceit about the innocuous elements of society. This technology can prove fatal not only to national security but on an international level. Existing methodologies that apply deep learning to automatically extract salient and discriminative features to detect Deepfakes based on typical CNN-LSTM models tend to have their shortcomings. Having said that, we propose a system that extracts Spatio-Temporal features and achieves Real-Time Deepfake detection using Transformers. For the end user, a web application was developed, which with utmost simplicity allows the uploading of a video that will be further authenticated within the application and, at the same time, features the authentication of live meetings.