Towards Generalizable Deepfake Video Detection via Saliency-guided and Temporal-Frequency Fusion Network

The proliferation of deepfake technology poses significant threats to information security, social trust, and personal privacy. Existing deepfake detection methods predominantly focus on spatial domain features and often overlook temporal inconsistencies and frequency cues. Thus, their robustness and generalization capabilities are limited, especially in deepfake video detection. To address these limitations, this paper proposes a Saliency-guided and Temporal-Frequency Fusion Network (STFFNet) for deepfake video detection. The proposed method integrates three key components: a Saliency-Guided Feature Perturbation (SGFP) module to bolster robustness, a Temporal Consistency Enhancement Transformer (TCET) optimized with contrastive loss to capture fine-grained inter-frame relations, and a Spatial-Frequency Fusion (SFF) module designed to synergistically fuse spatial and frequency signals for enhanced discriminative power. Comprehensive experiments on benchmark datasets (FF++, Celeb-DF-v1, Celeb-DF-v2, DFDCP, and FaceShifter) demonstrate that STFFNet achieves state-of-the-art performance with an average AUC of 92% in the frame level and 96% in the video level. Ablation studies validate the complementary contributions of each module, establishing STFFNet as a robust and generalizable solution for deepfake detection. The code is available at https://github.com/swc-code-cc/STFFNet .

Authors

Institutions

Publication Details

Journal
ACM Transactions on Multimedia Computing Communications and Applications
Published
2026-11-16
DOI
https://doi.org/10.1145/3860637
Primary Topic
Digital Media Forensic Detection
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Towards Generalizable Deepfake Video Detection via Saliency-guided and Temporal-Frequency Fusion Network

Mingliang Gao, Siyou Guo, Weicheng Song
ACM Transactions on Multimedia Computing Communications and Applications
Digital Media Forensic Detection
article

Towards Generalizable Deepfake Video Detection via Saliency-guided and Temporal-Frequency Fusion Network

Mingliang Gao, Siyou Guo, Weicheng Song
article en

Abstract

The proliferation of deepfake technology poses significant threats to information security, social trust, and personal privacy. Existing deepfake detection methods predominantly focus on spatial domain features and often overlook temporal inconsistencies and frequency cues. Thus, their robustness and generalization capabilities are limited, especially in deepfake video detection. To address these limitations, this paper proposes a Saliency-guided and Temporal-Frequency Fusion Network (STFFNet) for deepfake video detection. The proposed method integrates three key components: a Saliency-Guided Feature Perturbation (SGFP) module to bolster robustness, a Temporal Consistency Enhancement Transformer (TCET) optimized with contrastive loss to capture fine-grained inter-frame relations, and a Spatial-Frequency Fusion (SFF) module designed to synergistically fuse spatial and frequency signals for enhanced discriminative power. Comprehensive experiments on benchmark datasets (FF++, Celeb-DF-v1, Celeb-DF-v2, DFDCP, and FaceShifter) demonstrate that STFFNet achieves state-of-the-art performance with an average AUC of 92% in the frame level and 96% in the video level. Ablation studies validate the complementary contributions of each module, establishing STFFNet as a robust and generalizable solution for deepfake detection. The code is available at https://github.com/swc-code-cc/STFFNet .

ACM Transactions on Multimedia Computing Communications and Applications
Shandong University of Technology (CN)
Digital Media Forensic Detection
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon