FakeAVCeleb
收藏资源简介:
FakeAVCeleb数据集是一个独特的音频-视频多模态深度伪造数据集,由成均馆大学计算与信息学院创建。该数据集包含600个视频,涵盖五种不同种族和两个性别,确保了性别和种族的平衡。每个视频平均时长7秒,内容包括真实的视频但伪造的音频、真实的音频但伪造的视频以及音频和视频均为伪造的情况。数据集的创建过程中使用了多种深度伪造生成方法,包括面部交换和面部重演技术。FakeAVCeleb数据集主要用于研究和开发能够同时检测视频和音频深度伪造的检测器,以应对日益增长的深度伪造技术带来的安全挑战。
The FakeAVCeleb dataset is a unique audio-visual multimodal deepfake dataset created by the School of Computing and Information at Sungkyunkwan University. This dataset contains 600 videos covering five distinct ethnic groups and two genders, ensuring a balanced distribution across gender and ethnicity. Each video has an average duration of 7 seconds, with content covering three scenarios: real video paired with forged audio, real audio paired with forged video, and both audio and video being forged. A variety of deepfake generation methods were used during the dataset's creation, including facial swapping and facial reenactment techniques. The FakeAVCeleb dataset is primarily used for researching and developing detectors capable of simultaneously detecting both video and audio deepfakes, to address the escalating security challenges brought about by the growing prevalence of deepfake technologies.




