Authors: Gil Knafo, Ohad Fried

Published on: December 01, 2022

Impact Score: 8.22

Arxiv code: Arxiv:2212.00773

Summary

• What is new: The introduction of FakeOut, a deepfake detection method that leverages multi-modal, out-of-domain data in a self-supervised manner for better detection of unseen manipulations.
• Why this is important: The rise of synthetic videos in social media allowing the spread of misinformation and the challenge in detecting forgeries not seen during training.
• What the research proposes: FakeOut, a novel approach utilizing a multi-modal, out-of-domain backbone, trained self-supervised and adapted to video deepfake detection.
• Results: FakeOut achieves state-of-the-art results in detecting various types of deepfakes, especially those not encountered during training, across audio-visual datasets.

Technical Details

Technological frameworks used: Self-supervised learning framework adapted for multi-modal data

Models used: Custom FakeOut model

Data used: Audio-Visual datasets, including out-of-domain videos

Potential Impact

Social media platforms, news organizations, cybersecurity firms, and any entity relying on authenticity of video content

Want to implement this idea in a business?

We have generated a startup concept here: FakeOutTech.