A3:Ambiguous Aberrations Captured via Astray-Learning for Facial Forgery Semantic Sublimation

View on arXiv ← Back to list

Authors: Xinan He, Yue Zhou, Wei Ye, Feng Ding

Published: 2024-05-24 03:12:57+00:00

AI Summary

This paper introduces astray-learning, a novel approach for deepfake detection that enhances generalizability and fairness by blending hybrid forgery semantics into authentic imagery, creating ambiguous aberrations that reduce model bias.

Abstract

Prior DeepFake detection methods have faced a core challenge in preserving generalizability and fairness effectively. In this paper, we proposed an approach akin to decoupling and sublimating forgery semantics, named astray-learning. The primary objective of the proposed method is to blend hybrid forgery semantics derived from high-frequency components into authentic imagery, named aberrations. The ambiguity of aberrations is beneficial to reducing the model's bias towards specific semantics. Consequently, it can enhance the model's generalization ability and maintain the detection fairness. All codes for astray-learning are publicly available at https://anonymous.4open.science/r/astray-learning-C49B .

Key findings

The proposed astray-learning method significantly improves the average accuracy of deepfake detection by 6% across multiple datasets. It also enhances fairness and robustness compared to existing methods, as demonstrated through various evaluation metrics including AUC, FMAG, FFPR, and FMEO.

Approach

Astray-learning blends high-frequency forgery semantics (aberrations) from fake images into real images. This reduces model bias towards specific forgery techniques, improving generalization and fairness. A novel training strategy with astray loss and a feature fusion module prevents overfitting.

Datasets

FaceForensics++ (FF++), DeepFakeDetection (DFD), Deepfake Detection Challenge (DFDC), Celeb-DF

Model(s)

Xception, ResNet-34, ResNet-50, EfficientNet-b4

Author countries

China

← Previous