Attending Generalizability in Course of Deep Fake Detection by Exploring Multi-task Learning

TL;DR

This paper investigates multi-task learning and contrastive techniques to improve the generalizability of deep fake detection models across different manipulation methods, demonstrating superior performance on the FaceForensics++ dataset.

Contribution

It introduces a multi-task learning approach combined with contrastive techniques to enhance deep fake detection generalization across unseen manipulation methods.

Findings

The proposed model accurately detects unseen manipulation techniques.

Multi-task learning improves generalization over single-task models.

Contrastive techniques further enhance detection robustness.

Abstract

This work explores various ways of exploring multi-task learning (MTL) techniques aimed at classifying videos as original or manipulated in cross-manipulation scenario to attend generalizability in deep fake scenario. The dataset used in our evaluation is FaceForensics++, which features 1000 original videos manipulated by four different techniques, with a total of 5000 videos. We conduct extensive experiments on multi-task learning and contrastive techniques, which are well studied in literature for their generalization benefits. It can be concluded that the proposed detection model is quite generalized, i.e., accurately detects manipulation methods not encountered during training as compared to the state-of-the-art.