DATA: Multi-Disentanglement based Contrastive Learning for Open-World Semi-Supervised Deepfake Attribution

View on arXiv ← Back to list

Authors: Ming-Hui Liu, Xiao-Qian Liu, Xin Luo, Xin-Shun Xu

Published: 2025-05-07 13:05:32+00:00

AI Summary

This paper introduces DATA, a multi-disentanglement based contrastive learning framework for open-world semi-supervised deepfake attribution. DATA disentangles method-specific features from forgery-irrelevant information and uses an augmented-memory mechanism for novel class discovery, improving generalization and achieving state-of-the-art performance.

Abstract

Deepfake attribution (DFA) aims to perform multiclassification on different facial manipulation techniques, thereby mitigating the detrimental effects of forgery content on the social order and personal reputations. However, previous methods focus only on method-specific clues, which easily lead to overfitting, while overlooking the crucial role of common forgery features. Additionally, they struggle to distinguish between uncertain novel classes in more practical open-world scenarios. To address these issues, in this paper we propose an innovative multi-DisentAnglement based conTrastive leArning framework, DATA, to enhance the generalization ability on novel classes for the open-world semi-supervised deepfake attribution (OSS-DFA) task. Specifically, since all generation techniques can be abstracted into a similar architecture, DATA defines the concept of 'Orthonormal Deepfake Basis' for the first time and utilizes it to disentangle method-specific features, thereby reducing the overfitting on forgery-irrelevant information. Furthermore, an augmented-memory mechanism is designed to assist in novel class discovery and contrastive learning, which aims to obtain clear class boundaries for the novel classes through instance-level disentanglements. Additionally, to enhance the standardization and discrimination of features, DATA uses bases contrastive loss and center contrastive loss as auxiliaries for the aforementioned modules. Extensive experimental evaluations show that DATA achieves state-of-the-art performance on the OSS-DFA benchmark, e.g., there are notable accuracy improvements in 2.55% / 5.7% under different settings, compared with the existing methods.

Key findings

DATA achieves state-of-the-art performance on the OSS-DFA benchmark, showing notable accuracy improvements (2.55% to 5.7%) over existing methods. It demonstrates superior generalization ability, especially on novel and real classes, and robustness to low-quality data.

Approach

DATA uses a multi-disentanglement approach, defining 'Orthonormal Deepfake Basis' to disentangle method-specific features and reduce overfitting. An augmented-memory mechanism assists in novel class discovery and contrastive learning, improving class boundary definition for novel classes.

Datasets

OSS-DFA and OW-DFA++ datasets with various protocols (Protocol-1 to Protocol-6). These datasets include various deepfake generation methods like FaceSwap, DeepFaceLab, StyleGAN, and others.

Model(s)

ResNet-50 (pre-trained) as feature extractor; MLP for feature disentanglement; DBSCAN for clustering; Gram-Schmidt process for orthogonalization.

Author countries

China

← Previous