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

Camilo Luciano Fosco, Emilie L Josephs, Alex J Andonian, Aude Oliva

Primary Area: representation learning for computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: deepfakes, video understanding, video forensics, video manipulation

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: We introduce a state-of-the-art model to detect deepfakes that can additionally generate Deepfake Caricatures, a novel visual indicator that is stronger at evidencing fakes than textual indicators.

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.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 2227