FakeRadar: Probing Forgery Outliers to Detect Unknown Deepfake Videos

View on arXiv ← Back to list

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 focusing on improving cross-domain generalization against unknown manipulation techniques. It uses Forgery Outlier Probing (FOP) to synthesize virtual outlier samples in the feature space, simulating unseen forgeries by modeling dynamic subclusters. These outliers are then used in an Outlier-Guided Tri-Training (OGTT) process to optimize the model to distinguish real, fake, and outlier classes.

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 outperformed existing state-of-the-art algorithms across various deepfake benchmarks, particularly excelling in cross-domain evaluations. In cross-dataset testing, the method achieved the best AUC across all evaluated external datasets (CDFv2, DFDCP, DFDC, DFD), demonstrating enhanced generalization to unseen manipulation techniques. Ablation studies confirmed that both Forgery Outlier Probing and Outlier-Guided Tri-Training were critical for achieving these substantial performance gains.

Approach

The approach involves Forgery Outlier Probing (FOP) where dynamic subcluster modeling identifies fine-grained forgery patterns using GMMs, followed by cluster-conditional outlier generation to synthesize novel forgery artifacts near subcluster boundaries. An Outlier-Guided Tri-Training (OGTT) strategy optimizes a triplet-class classifier using contrastive and cross-entropy losses to explicitly differentiate real, known fake, and these generated outlier samples.

Datasets

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

Model(s)

CLIP ViT-B/16 (Vision Transformer) with ST-Adapter

Author countries

China

← Previous