AV-Deepfake1M
收藏资源简介:
AV-Deepfake1M是由莫纳什大学等机构创建的大型音频-视觉深度伪造数据集,包含超过114万个视频,涉及2068个独特主题。该数据集通过大型语言模型生成,采用多种音频-视觉内容操纵策略,旨在推动时间深度伪造定位技术的研究。数据集内容包括视频操纵、音频操纵和音频-视觉操纵,适用于开发下一代深度伪造定位方法,以应对高度真实的深度伪造内容检测和定位挑战。
AV-Deepfake1M is a large-scale audio-visual deepfake dataset developed by Monash University and other institutions, which contains over 1.14 million videos covering 2068 unique subjects. Generated using large language models and leveraging multiple audio-visual content manipulation strategies, this dataset aims to advance research on temporal deepfake localization technology. The dataset covers video manipulation, audio manipulation, and audio-visual manipulation scenarios, and is suitable for developing next-generation deepfake localization methods to tackle the challenges of detecting and localizing highly realistic deepfake content.

- 1AV-Deepfake1M: A Large-Scale LLM-Driven Audio-Visual Deepfake Dataset莫纳什大学 · 2023年



