Decoupling Forgery Semantics for Generalizable Deepfake Detection

TL;DR

This paper introduces a semantic decoupling approach to improve the generalization of DeepFake detection models by extracting common forgery semantics and employing specialized training strategies.

Contribution

It proposes a novel semantic decoupling method combined with an adaptive high-pass module and two-stage training to enhance DeepFake detector generalizability.

Findings

Achieves superior detection accuracy across multiple datasets.

Demonstrates improved generalization to unseen DeepFake forgeries.

Outperforms existing methods in robustness and accuracy.

Abstract

In this paper, we propose a novel method for detecting DeepFakes, enhancing the generalization of detection through semantic decoupling. There are now multiple DeepFake forgery technologies that not only possess unique forgery semantics but may also share common forgery semantics. The unique forgery semantics and irrelevant content semantics may promote over-fitting and hamper generalization for DeepFake detectors. For our proposed method, after decoupling, the common forgery semantics could be extracted from DeepFakes, and subsequently be employed for developing the generalizability of DeepFake detectors. Also, to pursue additional generalizability, we designed an adaptive high-pass module and a two-stage training strategy to improve the independence of decoupled semantics. Evaluation on FF++, Celeb-DF, DFD, and DFDC datasets showcases our method's excellent detection and generalization performance. Code is available at: https://github.com/leaffeall/DFS-GDD.