Deepfake Forensics Adapter: A Dual-Stream Network for Generalizable Deepfake Detection

Jianfeng Liao; Yichen Wei; Raymond Chan Ching Bon; Shulan Wang; Kam-Pui Chow; Kwok-Yan Lam

cs.CVMar 2, 2026

Deepfake Forensics Adapter: A Dual-Stream Network for Generalizable Deepfake Detection

Authors: Jianfeng Liao, Yichen Wei, Raymond Chan Ching Bon, Shulan Wang, Kam-Pui Chow, Kwok-Yan Lam

Organizations: Shenzhen Technology University, Guangdong, China · Singapore Institute of Technology, Singapore, Singapore · The University of Hong Kong, Hong Kong, China · Nanyang Technological University, Singapore, Singapore

Abstract

The rapid advancement of deepfake generation techniques poses significant threats to public safety and causes societal harm through the creation of highly realistic synthetic facial media. While existing detection methods demonstrate limitations in generalizing to emerging forgery patterns, this paper presents Deepfake Forensics Adapter (DFA), a novel dual-stream framework that synergizes vision-language foundation models with targeted forensics analysis. Our approach integrates a pre-trained CLIP model with three core components to achieve specialized deepfake detection by leveraging the powerful general capabilities of CLIP without changing CLIP parameters: 1) A Global Feature Adapter is used to identify global inconsistencies in image content that may indicate forgery, 2) A Local Anomaly Stream enhances the model's ability to perceive local facial forgery cues by explicitly leveraging facial structure priors, and 3) An Interactive Fusion Classifier promotes deep interaction and fusion between global and local features using a transformer encoder. Extensive evaluations of frame-level and video-level benchmarks demonstrate the superior generalization capabilities of DFA, particularly achieving state-of-the-art performance in the challenging DFDC dataset with frame-level AUC/EER of 0.816/0.256 and video-level AUC/EER of 0.836/0.251, representing a 4.8% video AUC improvement over previous methods. Our framework not only demonstrates state-of-the-art performance, but also points out a feasible and effective direction for developing a robust deepfake detection system with enhanced generalization capabilities against the evolving deepfake threats. Our code is available at https://github.com/Liao330/DFA.git

Figures & tables

Figure 1: The radar chart shows how different models perform across several metrics.

Figure 2: Examples of reenactment, replacement, editing, and synthesis of deepfakes of the human face. [ 25 ]

Figure 3: Overall architecture of the proposed Deepfake Forensics Adapter.

Figure 4: Overview of datasets used for evaluating model performance, including CelebDF-V1, CelebDF-V2, DFDC, DFDCP, and FF++. DFDC is reserved for evaluating generalization capability, while the remaining datasets form the training/validation data.

Method	Acc	Precision	AUC	Avg
CNN [ 30 ]	0.904	0.815	0.890	0.867
Xception [ 12 ]	0.955	0.897	0.962	0.938
ResNeXt-LSTM [ 14 ]	0.967	0.931	0.960	0.953
CViT [ 13 ]	0.941	0.868	0.959	0.923
Efficient-ViT [ 15 ]	0.968	0.932	0.954	0.951
Ours	0.983	0.963	0.976	0.974

Table 1: Frame-level metrics scores of various deepfake detection methods on our Mixed dataset.

Method	AUC	EER $\downarrow$
CNN [ 30 ]	0.674	0.370
Xception [ 12 ]	0.752	0.315
ResNeXt-LSTM [ 14 ]	0.754	0.311
CViT [ 13 ]	0.738	0.323
Efficient-ViT [ 15 ]	0.764	0.293
Ours	0.816	0.256

Table 2: Frame-level AUC and EER scores of various deepfake detection methods on the DFDC dataset.

Method	AUC	EER $\downarrow$
CNN [ 30 ]	0.698	0.375
Xception [ 12 ]	0.774	0.325
ResNeXt-LSTM [ 14 ]	0.776	0.319
CViT [ 13 ]	0.770	0.323
Efficient-ViT [ 15 ]	0.788	0.300
Ours	0.836	0.251

Table 3: Video-level AUC and EER scores of various deepfake detection methods on the DFDC dataset.

Global	Local	IFC	AUC	EER $\downarrow$
$\times$	$\checkmark$	$\checkmark$	0.766	0.356
$\checkmark$	$\times$	$\checkmark$	0.747	0.323
$\checkmark$	$\checkmark$	$\times$	0.753	0.301
$\checkmark$	$\checkmark$	$\checkmark$	0.816	0.256

Table 4: Ablation study on DFA modules: Frame-level results on the DFDC dataset.

Figure 5: Comparison of t-SNE visualizations of features extracted by Xception (left) and our DFA model (right) on samples from the DFDC dataset.

Explore similar work

GazeCLIP: Gaze-Guided CLIP with Adaptive-Enhanced Fine-Grained Language Prompt for Deepfake Attribution and Detection
Mar 31, 2026Yaning Zhang, Linlin Shen, Zitong Yu +2Deepfake Detection Digital Forensics
Faculty of Computer Science and Technology, Qilu University of Technology (Shandong Academy of Sciences), Jinan, 250014, China · Computer Vision Institute, College of Computer Science and Software Engineering, Shenzhen University, Shenzhen, 518060, China · National Engineering Laboratory for Big Data System Computing Technology, Shenzhen University +5
Cross-Domain Generalization Limits of Vision Foundation Models in Facial Deepfake Detection
May 24, 2026Ibrahim DelibasogluVision Foundation Models Cross-Domain Generalization
Department of Software Engineering, Faculty of Computer and Information Sciences, Sakarya University, Esentepe, Sakarya, 54050 Türkiye.
Multi-Agent Forensic Reasoning for Generalizable Deepfake Video Detection
Aug 7, 2026Xuechao Zou, Shun Zhang, Kai Li +6Explainable Deepfake Detection Deepfake Detection
Beijing Jiaotong University · Tsinghua University · Ant Group

Mar 31, 2026J/K move · Enter open · S save

GazeCLIP: Gaze-Guided CLIP with Adaptive-Enhanced Fine-Grained Language Prompt for Deepfake Attribution and Detection

Yaning Zhang, Linlin Shen, Zitong Yu, Chunjie Ma+1

Faculty of Computer Science and Technology, Qilu University of Technology (Shandong Academy of Sciences), Jinan, 250014, China · Computer Vision Institute, College of Computer Science and Software Engineering, Shenzhen University, Shenzhen, 518060, China · National Engineering Laboratory for Big Data System Computing Technology, Shenzhen University+5

Deepfake Detection Digital Forensics VLM Adaptation Vision-Language Pretraining

Open paper AskPDF arXiv

Current deepfake attribution or deepfake detection works tend to exhibit poor generalization to novel generative methods due to the limited exploration in visual modalities alone. They tend to assess the attribution or detection performance of models on unseen advanced generators, coarsely, and fail to consider the synergy of the two tasks. To this end, we propose a novel gaze-guided CLIP with adaptive-enhanced fine-grained language prompts for fine-grained deepfake attribution and detection (DFAD). Specifically, we conduct a novel and fine-grained benchmark to evaluate the DFAD performance of networks on novel generators like diffusion and flow models. Additionally, we introduce a gaze-aware model based on CLIP, which is devised to enhance the generalization to unseen face forgery attacks. Built upon the novel observation that there are significant distribution differences between pristine and forged gaze vectors, and the preservation of the target gaze in facial images generated by GAN and diffusion varies significantly, we design a visual perception encoder to employ the inherent gaze differences to mine global forgery embeddings across appearance and gaze domains. We propose a gaze-aware image encoder (GIE) that fuses forgery gaze prompts extracted via a gaze encoder with common forged image embeddings to capture general attribution patterns, allowing features to be transformed into a more stable and common DFAD feature space. We build a language refinement encoder (LRE) to generate dynamically enhanced language embeddings via an adaptive-enhanced word selector for precise vision-language matching. Extensive experiments on our benchmark show that our model outperforms the state-of-the-art by 6.56% ACC and 5.32% AUC in average performance under the attribution and detection settings, respectively. Codes will be available on GitHub.

Abstract

Explore similar work

GazeCLIP: Gaze-Guided CLIP with Adaptive-Enhanced Fine-Grained Language Prompt for Deepfake Attribution and Detection

Cross-Domain Generalization Limits of Vision Foundation Models in Facial Deepfake Detection

Multi-Agent Forensic Reasoning for Generalizable Deepfake Video Detection