Deepfake Caricatures: Amplifying attention to artifacts increases deepfake detection by humans and machines

View on arXiv ← Back to list

Authors: Camilo Fosco, Emilie Josephs, Alex Andonian, Aude Oliva

Published: 2022-06-01 14:43:49+00:00

AI Summary

This paper introduces Deepfake Caricatures, a framework that amplifies artifacts in deepfake videos to improve human and machine detection. It proposes a novel Artifact Attention module trained on human responses to highlight and magnify video artifacts, creating easily detectable visual indicators. A deepfake detection model incorporating this module shows improved accuracy and robustness.

Abstract

Deepfakes can fuel online misinformation. As deepfakes get harder to recognize with the naked eye, human users become more reliant on deepfake detection models to help them decide whether a video is real or fake. Currently, models yield a prediction for a video's authenticity, but do not integrate a method for alerting a human user. We introduce a framework for amplifying artifacts in deepfake videos to make them more detectable by people. We propose a novel, semi-supervised Artifact Attention module, which is trained on human responses to create attention maps that highlight video artifacts, and magnify them to create a novel visual indicator we call Deepfake Caricatures. In a user study, we demonstrate that Caricatures greatly increase human detection, across video presentation times and user engagement levels. We also introduce a deepfake detection model that incorporates the Artifact Attention module to increase its accuracy and robustness. Overall, we demonstrate the success of a human-centered approach to designing deepfake mitigation methods.

Key findings

Deepfake Caricatures significantly increased human detection accuracy, even with short viewing times and varying levels of user engagement. The deepfake detection model incorporating the Artifact Attention module achieved near state-of-the-art performance across multiple datasets. The method showed robustness to various video perturbations.

Approach

The authors propose a framework that uses a semi-supervised Artifact Attention module trained on human annotations of deepfake artifacts. This module generates attention maps highlighting artifacts, which are then magnified by a Caricature Generation Module to create Deepfake Caricatures for improved human detection. A deepfake detection model is also developed that incorporates the Artifact Attention module.

Datasets

DeepFake Detection Challenge Dataset preview (DFDCp), FaceForensics++, CelebDFv2, DeeperForensics, FaceShifter

Model(s)

CariNet (consisting of an Artifact Attention Module, a Classifier Module using EVA-02 backbone and self-attention blocks modulated by artifact heatmaps, and a Caricature Generation Module), Xception, ResNet

Author countries

USA

← Previous