A Sanity Check for Multi-In-Domain Face Forgery Detection in the Real World

Authors: Jikang Cheng, Renye Yan, Zhiyuan Yan, Yaozhong Gan, Xueyi Zhang, Zhongyuan Wang, Wei Peng, Ling Liang

Published: 2025-12-04 14:21:08+00:00

AI Summary

This paper introduces Multi-In-Domain Face Forgery Detection (MID-FFD) to better reflect real-world deepfake scenarios with diverse and extensive forgery domains. It proposes DevDet, a model-agnostic framework comprising a Face Forgery Developer (FFDev) and a Dose-Adaptive detector Fine-Tuning (DAFT) strategy. DevDet amplifies real/fake differences to dominate the feature space over domain discrepancies, significantly improving detection accuracy in MID-FFD while maintaining generalization to unseen data.

Abstract

Existing methods for deepfake detection aim to develop generalizable detectors. Although generalizable is the ultimate target once and for all, with limited training forgeries and domains, it appears idealistic to expect generalization that covers entirely unseen variations, especially given the diversity of real-world deepfakes. Therefore, introducing large-scale multi-domain data for training can be feasible and important for real-world applications. However, within such a multi-domain scenario, the differences between multiple domains, rather than the subtle real/fake distinctions, dominate the feature space. As a result, despite detectors being able to relatively separate real and fake within each domain (i.e., high AUC), they struggle with single-image real/fake judgments in domain-unspecified conditions (i.e., low ACC). In this paper, we first define a new research paradigm named Multi-In-Domain Face Forgery Detection (MID-FFD), which includes sufficient volumes of real-fake domains for training. Then, the detector should provide definitive real-fake judgments to the domain-unspecified inputs, which simulate the frame-by-frame independent detection scenario in the real world. Meanwhile, to address the domain-dominant issue, we propose a model-agnostic framework termed DevDet (Developer for Detector) to amplify real/fake differences and make them dominant in the feature space. DevDet consists of a Face Forgery Developer (FFDev) and a Dose-Adaptive detector Fine-Tuning strategy (DAFT). Experiments demonstrate our superiority in predicting real-fake under the MID-FFD scenario while maintaining original generalization ability to unseen data.

Key findings

Existing detectors struggle with domain-unspecified, frame-by-frame real/fake judgments in multi-domain scenarios due to domain differences dominating the feature space. DevDet significantly enhances detection confidence, achieving up to an 11.80% improvement in MID-FFD performance. The framework effectively preserves the original generalization ability of pretrained detectors to out-of-domain data.

Approach

The authors propose DevDet, a two-stage model-agnostic framework. Stage 1 involves a Face Forgery Developer (FFDev) that acts as a preprocessing step to amplify forgery traces, trained on easy-real and hard-fake samples. Stage 2 employs a Dose-Adaptive detector Fine-Tuning (DAFT) strategy, which uses a DoseDict to dynamically adjust the FFDev dose for inputs, enabling the detector to adapt to developed images while preserving its original generalization capability.

Datasets

Celeb-DF-v2 (CDF), DeepFake Detection Challenge Preview (DFDCP), FaceForensics++ (FF++), WildDeepfake (WDF), DiffusionFace (DiffFace), DF40, Celeb-DF++ (CDF3). These datasets include forgeries from methods like BlendFace, Simswap, DiT, SiT, AniTalker, FLOAT, DDIM, and DiffSwap.

Model(s)

Xception, EffNet-B4 (Effnb4), Capsule, CLIP, F3Net, SPSL, SBI, IID, ProDet, Effort (used as the base model for main comparison).

Author countries

China, USA

← Previous