Generalizing Video DeepFake Detection by Self-generated Audio-Visual Pseudo-Fakes

Authors: Zihe Wei, Yuezun Li

Published: 2026-04-10 08:42:31+00:00

AI Summary

This paper introduces AVPF (Audio-Visual Pseudo-Fakes), a novel method to significantly improve the generalizability of video deepfake detection. AVPF generates diverse pseudo-fake training samples solely from authentic videos by simulating common audio-visual correspondence patterns found in real deepfakes. This approach, which requires no actual deepfake samples for training, leads to an average performance improvement of up to 7.4% across multiple standard datasets.

Abstract

Detecting video deepfakes has become increasingly urgent in recent years. Given the audio-visual information in videos, existing methods typically expose deepfakes by modeling cross-modal correspondence using specifically designed architectures with publicly available datasets. While they have shown promising results, their effectiveness often degrades in real-world scenarios, as the limited diversity of training datasets naturally restricts generalizability to unseen cases. To address this, we propose a simple yet effective method, called AVPF, which can notably enhance model generalizability by training with self-generated Audio-Visual Pseudo-Fakes.The key idea of AVPF is to create pseudo-fake training samples that contain diverse audio-visual correspondence patterns commonly observed in real-world deepfakes. We highlight that AVPF is generated solely from authentic samples, and training relies only on authentic data and AVPF, without requiring any real deepfakes.Extensive experiments on multiple standard datasets demonstrate the strong generalizability of the proposed method, achieving an average performance improvement of up to 7.4%.