Deepfake Detection Method Integrating Multiple Parameter-Efficient Fine-Tuning Techniques

In recent years, as deepfake technology matures, face-swapping software and synthesized videos have become widespread. While these techniques offer entertainment, they also provide opportunities for misuse by malicious actors. Consequently, the significance of deepfake detection technology has grown...

Full description

Saved in:
Bibliographic Details
Main Author: ZHANG Yiwen, CAI Manchun, CHEN Yonghao, ZHU Yi, YAO Lifeng
Format: Article
Language:zho
Published: Journal of Computer Engineering and Applications Beijing Co., Ltd., Science Press 2024-12-01
Series:Jisuanji kexue yu tansuo
Subjects:
Online Access:http://fcst.ceaj.org/fileup/1673-9418/PDF/2311053.pdf
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850057369547964416
author ZHANG Yiwen, CAI Manchun, CHEN Yonghao, ZHU Yi, YAO Lifeng
author_facet ZHANG Yiwen, CAI Manchun, CHEN Yonghao, ZHU Yi, YAO Lifeng
author_sort ZHANG Yiwen, CAI Manchun, CHEN Yonghao, ZHU Yi, YAO Lifeng
collection DOAJ
description In recent years, as deepfake technology matures, face-swapping software and synthesized videos have become widespread. While these techniques offer entertainment, they also provide opportunities for misuse by malicious actors. Consequently, the significance of deepfake detection technology has grown markedly. Existing methods for deepfake detection commonly suffer from issues including poor cross-compression robustness, weak cross-dataset generalization, and high model training overheads. To address these challenges, this paper proposes a deepfake detection approach that combines multiple parameter-efficient fine-tuning techniques. This method utilizes a visual Transformer model pretrained with the masked image modeling self-supervised method as its backbone. Initially, it employs the low-rank adaptation (LoRA) method for fine-tuning the self-attention module parameters of the pretrained model. Concurrently, it introduces a parallel structure incorporating convolutional adapters to capture local texture information, enhancing the model’s adaptability in deepfake detection tasks. Subsequently, a serial structure introduces classical adapters to fine-tune the feed-forward network of the pretrained model, maximizing the utilization of knowledge acquired during the pretraining phase. Ultimately, a multi-layer perceptron replaces the original pretrained model’s classification head for deepfake detection. Experimental results across six datasets demonstrate that this model achieves an average frame-level AUC of approximately 0.996 with only 2×107 trainable parameters. In cross-compression experiments, the average frame-level AUC drop is 0.135. In cross-dataset generalization experiments, the frame-level AUC averages around 0.765.
format Article
id doaj-art-8d514d8a8dc24b9e805183d2f7c2c93d
institution DOAJ
issn 1673-9418
language zho
publishDate 2024-12-01
publisher Journal of Computer Engineering and Applications Beijing Co., Ltd., Science Press
record_format Article
series Jisuanji kexue yu tansuo
spelling doaj-art-8d514d8a8dc24b9e805183d2f7c2c93d2025-08-20T02:51:27ZzhoJournal of Computer Engineering and Applications Beijing Co., Ltd., Science PressJisuanji kexue yu tansuo1673-94182024-12-0118123335334710.3778/j.issn.1673-9418.2311053Deepfake Detection Method Integrating Multiple Parameter-Efficient Fine-Tuning TechniquesZHANG Yiwen, CAI Manchun, CHEN Yonghao, ZHU Yi, YAO Lifeng0College of Information and Cyber Security, People’s Public Security University of China, Beijing 100038, ChinaIn recent years, as deepfake technology matures, face-swapping software and synthesized videos have become widespread. While these techniques offer entertainment, they also provide opportunities for misuse by malicious actors. Consequently, the significance of deepfake detection technology has grown markedly. Existing methods for deepfake detection commonly suffer from issues including poor cross-compression robustness, weak cross-dataset generalization, and high model training overheads. To address these challenges, this paper proposes a deepfake detection approach that combines multiple parameter-efficient fine-tuning techniques. This method utilizes a visual Transformer model pretrained with the masked image modeling self-supervised method as its backbone. Initially, it employs the low-rank adaptation (LoRA) method for fine-tuning the self-attention module parameters of the pretrained model. Concurrently, it introduces a parallel structure incorporating convolutional adapters to capture local texture information, enhancing the model’s adaptability in deepfake detection tasks. Subsequently, a serial structure introduces classical adapters to fine-tune the feed-forward network of the pretrained model, maximizing the utilization of knowledge acquired during the pretraining phase. Ultimately, a multi-layer perceptron replaces the original pretrained model’s classification head for deepfake detection. Experimental results across six datasets demonstrate that this model achieves an average frame-level AUC of approximately 0.996 with only 2×107 trainable parameters. In cross-compression experiments, the average frame-level AUC drop is 0.135. In cross-dataset generalization experiments, the frame-level AUC averages around 0.765.http://fcst.ceaj.org/fileup/1673-9418/PDF/2311053.pdfdeepfakes; vision transformer; self-supervised pretrained models; low-rank adaptation (lora); parameter-efficient fine-tuning
spellingShingle ZHANG Yiwen, CAI Manchun, CHEN Yonghao, ZHU Yi, YAO Lifeng
Deepfake Detection Method Integrating Multiple Parameter-Efficient Fine-Tuning Techniques
Jisuanji kexue yu tansuo
deepfakes; vision transformer; self-supervised pretrained models; low-rank adaptation (lora); parameter-efficient fine-tuning
title Deepfake Detection Method Integrating Multiple Parameter-Efficient Fine-Tuning Techniques
title_full Deepfake Detection Method Integrating Multiple Parameter-Efficient Fine-Tuning Techniques
title_fullStr Deepfake Detection Method Integrating Multiple Parameter-Efficient Fine-Tuning Techniques
title_full_unstemmed Deepfake Detection Method Integrating Multiple Parameter-Efficient Fine-Tuning Techniques
title_short Deepfake Detection Method Integrating Multiple Parameter-Efficient Fine-Tuning Techniques
title_sort deepfake detection method integrating multiple parameter efficient fine tuning techniques
topic deepfakes; vision transformer; self-supervised pretrained models; low-rank adaptation (lora); parameter-efficient fine-tuning
url http://fcst.ceaj.org/fileup/1673-9418/PDF/2311053.pdf
work_keys_str_mv AT zhangyiwencaimanchunchenyonghaozhuyiyaolifeng deepfakedetectionmethodintegratingmultipleparameterefficientfinetuningtechniques