Scholarlink

<p>Recent advancements in Deepfake technology have raised significant concerns regarding the authenticity of online media. Misuse of Deepfakes poses threats to privacy and economic security. To address the generalization issues faced by existing Deepfake detection methods, we propose a novel framework combining forgery-aware Dynamic Data Augmentation (DDA) and Cross-Frame Multi-Head Attention (CMA). DDA generates forgery heatmaps using spatial features extracted by Swin Transformer, guiding region-level data augmentation dynamically. CMA extracts temporal inconsistencies across frames using spatial features. Spatiotemporal features are fused and fed into a Bidirectional GRU (Bi-GRU) to model sequential dependencies. Extensive experiments demonstrate that our framework outperforms state-of-the-art methods in cross-compression and cross-forgery-type evaluations. Our code and datasets are publicly available at <a href="https://github.com/luanmianv/DDA-CMA.">https://github.com/luanmianv/DDA-CMA.</a></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

Enhanced deepfake detection via dynamic data augmentation and spatiotemporal attention

  • 期刊论文

  • 发表日期：

    2025年12月10日

查看全文
  • Teng Zhang,
  • Gen Li,
  • Yanhui Xiao,
  • Huawei Tian,
  • Yun Cao

摘要

Recent advancements in Deepfake technology have raised significant concerns regarding the authenticity of online media. Misuse of Deepfakes poses threats to privacy and economic security. To address the generalization issues faced by existing Deepfake detection methods, we propose a novel framework combining forgery-aware Dynamic Data Augmentation (DDA) and Cross-Frame Multi-Head Attention (CMA). DDA generates forgery heatmaps using spatial features extracted by Swin Transformer, guiding region-level data augmentation dynamically. CMA extracts temporal inconsistencies across frames using spatial features. Spatiotemporal features are fused and fed into a Bidirectional GRU (Bi-GRU) to model sequential dependencies. Extensive experiments demonstrate that our framework outperforms state-of-the-art methods in cross-compression and cross-forgery-type evaluations. Our code and datasets are publicly available at https://github.com/luanmianv/DDA-CMA.