Scholarlink

<p>Deepfakes have become a significant social and political issue due to their potential for misuse. While several deepfake detection techniques have been proposed, their performance is often limited by the kind of training data. This results in low accuracy on Middle Eastern and Indian deepfake content. To address this issue, a new dataset, IAV-DF (Indian Audio–Video DeepFake Dataset), is introduced, containing a diversified collection of 10,717 real and 2,47,678 deepfake videos synthesized using four distinct deepfaking procedures. The dataset also includes lip-sync deepfakes generated using the cloned voice of the target person, making them more realistic. The dataset captures diversified environmental settings, people of different ages, genders and facial features that addresses the diversity of Indian faces, including beard, colour, turban, etc. This paper provides a comprehensive evaluation of existing deepfake detection techniques on IAV-DF and other available datasets, highlighting the significance of diverse datasets for robust deepfake detection techniques. The dataset will be published through this link: <a href="https://github.com/Staffy15/IAV-DF">https://github.com/Staffy15/IAV-DF</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

Assessing deepfake detection methods: a comparative evaluation on novel large-scale Asian deepfake dataset

  • 期刊论文

  • 发表日期：

    2025年3月6日

查看全文
  • Staffy Kingra,
  • Naveen Aggarwal,
  • Nirmal Kaur

摘要

Deepfakes have become a significant social and political issue due to their potential for misuse. While several deepfake detection techniques have been proposed, their performance is often limited by the kind of training data. This results in low accuracy on Middle Eastern and Indian deepfake content. To address this issue, a new dataset, IAV-DF (Indian Audio–Video DeepFake Dataset), is introduced, containing a diversified collection of 10,717 real and 2,47,678 deepfake videos synthesized using four distinct deepfaking procedures. The dataset also includes lip-sync deepfakes generated using the cloned voice of the target person, making them more realistic. The dataset captures diversified environmental settings, people of different ages, genders and facial features that addresses the diversity of Indian faces, including beard, colour, turban, etc. This paper provides a comprehensive evaluation of existing deepfake detection techniques on IAV-DF and other available datasets, highlighting the significance of diverse datasets for robust deepfake detection techniques. The dataset will be published through this link: https://github.com/Staffy15/IAV-DF.