Scholarlink

<p>Deepfake detection models often fail to generalize across unseen manipulations and real-world video degradations. Despite numerous proposed methods, prior work lacks a systematic assessment of how data augmentation strategies affect forensic robustness. In this study, we perform a comprehensive evaluation of 14 data augmentation techniques for deepfake detection. Using the Xception backbone across FaceForensics++, DFDC-P, and Celeb-DF, we show that data augmentation improves both in-domain accuracy and cross-dataset generalization. In particular, frequency-aware strategies, such as FourierMix, substantially improve accuracy by up to +2–3% across multiple forgery types by amplifying spectral inconsistencies introduced during synthesis, while JPEG compression models real-world degradations and enhances decision confidence. Despite these gains, general-purpose, augmentation-enhanced models remain outperformed by specialized forensic architectures, such as Face X-ray and LipForensics by more than 10%, which explicitly leverage manipulation-specific cues. These findings provide practical guidance for designing model-agnostic augmentation pipelines and highlight the importance of hybrid approaches that combine effective data augmentations with specialized forensic architectures to achieve reliable and robust deepfake detection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Does data augmentation help or hinder the generalization of deepfake video detection?

  • 期刊论文

  • 发表日期：

    2026年3月4日

查看全文
  • Bachir Kaddar,
  • Sid Ahmed Fezza,
  • Elhocine Boutellaa,
  • Wassim Hamidouche,
  • Abdenour Hadid

摘要

Deepfake detection models often fail to generalize across unseen manipulations and real-world video degradations. Despite numerous proposed methods, prior work lacks a systematic assessment of how data augmentation strategies affect forensic robustness. In this study, we perform a comprehensive evaluation of 14 data augmentation techniques for deepfake detection. Using the Xception backbone across FaceForensics++, DFDC-P, and Celeb-DF, we show that data augmentation improves both in-domain accuracy and cross-dataset generalization. In particular, frequency-aware strategies, such as FourierMix, substantially improve accuracy by up to +2–3% across multiple forgery types by amplifying spectral inconsistencies introduced during synthesis, while JPEG compression models real-world degradations and enhances decision confidence. Despite these gains, general-purpose, augmentation-enhanced models remain outperformed by specialized forensic architectures, such as Face X-ray and LipForensics by more than 10%, which explicitly leverage manipulation-specific cues. These findings provide practical guidance for designing model-agnostic augmentation pipelines and highlight the importance of hybrid approaches that combine effective data augmentations with specialized forensic architectures to achieve reliable and robust deepfake detection.