FakeRadar: Probing Forgery Outliers to Detect Unknown Deepfake Videos

Authors: Zhaolun Li, Jichang Li, Yinqi Cai, Junye Chen, Xiaonan Luo, Guanbin Li, Rushi Lan

Published: 2025-12-16 17:11:45+00:00

AI Summary

FakeRadar is a novel deepfake video detection framework designed to improve cross-domain generalization against emerging manipulation techniques. It proactively probes for forgery outliers in the feature space using dynamic subcluster modeling and cluster-conditional outlier generation. The framework employs an Outlier-Guided Tri-Training strategy with outlier-driven contrastive and cross-entropy losses to optimize the detector, enabling it to distinguish real, known fake, and unseen outlier samples effectively.

Abstract

In this paper, we propose FakeRadar, a novel deepfake video detection framework designed to address the challenges of cross-domain generalization in real-world scenarios. Existing detection methods typically rely on manipulation-specific cues, performing well on known forgery types but exhibiting severe limitations against emerging manipulation techniques. This poor generalization stems from their inability to adapt effectively to unseen forgery patterns. To overcome this, we leverage large-scale pretrained models (e.g. CLIP) to proactively probe the feature space, explicitly highlighting distributional gaps between real videos, known forgeries, and unseen manipulations. Specifically, FakeRadar introduces Forgery Outlier Probing, which employs dynamic subcluster modeling and cluster-conditional outlier generation to synthesize outlier samples near boundaries of estimated subclusters, simulating novel forgery artifacts beyond known manipulation types. Additionally, we design Outlier-Guided Tri-Training, which optimizes the detector to distinguish real, fake, and outlier samples using proposed outlier-driven contrastive learning and outlier-conditioned cross-entropy losses. Experiments show that FakeRadar outperforms existing methods across various benchmark datasets for deepfake video detection, particularly in cross-domain evaluations, by handling the variety of emerging manipulation techniques.

Key findings

FakeRadar consistently outperforms existing state-of-the-art deepfake detection algorithms across various benchmark datasets, particularly excelling in cross-domain and cross-manipulation evaluations. It demonstrates superior generalization capacity, achieving significant AUC improvements (e.g., 3.6% and 3.7% over leading methods on DFDC). The proposed Forgery Outlier Probing and Outlier-Guided Tri-Training components are critical for enhancing detection robustness and feature discriminability against unseen forgery types.

Approach

FakeRadar leverages large-scale pre-trained models (CLIP) to identify distributional gaps between real, known fake, and unseen manipulations through 'Forgery Outlier Probing.' This involves dynamic subcluster modeling and cluster-conditional outlier generation to synthesize outlier samples simulating novel forgery artifacts. The detector is optimized using 'Outlier-Guided Tri-Training,' which employs outlier-driven contrastive learning and outlier-conditioned cross-entropy losses to distinguish between real, fake, and outlier samples.

Datasets

FaceForensics++ (FF++), CDFv2, DFDCP, DFDC, DFD

Model(s)

CLIP (ViT-B/16) as backbone, ST-Adapter layers, Triplet-class classifier (single linear layer with softmax activation)

Author countries

China

← Previous