Combating Digitally Altered Images: Deepfake Detection

TL;DR

This paper introduces a robust Deepfake detection method using a modified Vision Transformer trained on augmented datasets, achieving state-of-the-art accuracy in distinguishing real from manipulated images.

Contribution

It presents a novel Deepfake detection approach based on a modified Vision Transformer with data augmentation and class imbalance handling, improving detection robustness.

Findings

Achieved state-of-the-art accuracy on Deepfake detection

Effective handling of class imbalance in training data

Robust detection across diverse manipulated images

Abstract

The rise of Deepfake technology to generate hyper-realistic manipulated images and videos poses a significant challenge to the public and relevant authorities. This study presents a robust Deepfake detection based on a modified Vision Transformer(ViT) model, trained to distinguish between real and Deepfake images. The model has been trained on a subset of the OpenForensics Dataset with multiple augmentation techniques to increase robustness for diverse image manipulations. The class imbalance issues are handled by oversampling and a train-validation split of the dataset in a stratified manner. Performance is evaluated using the accuracy metric on the training and testing datasets, followed by a prediction score on a random image of people, irrespective of their realness. The model demonstrates state-of-the-art results on the test dataset to meticulously detect Deepfake images.