DeeperForensics-1.0
收藏资源简介:
DeeperForensics-1.0是由南洋理工大学和商汤科技研究院共同创建的大型人脸伪造检测数据集,包含60,000个视频,总计1760万帧,是现有同类数据集中规模最大的。该数据集通过应用广泛的现实世界扰动,以更高的规模和多样性创建更具挑战性的基准。所有源视频都经过精心收集,并通过一种新的端到端人脸交换框架生成伪造视频。数据集的生成视频质量优于现有数据集,并通过用户研究得到验证。DeeperForensics-1.0还包含一个隐藏测试集,其中包含在人类评估中获得高分欺骗性的操纵视频。该数据集旨在推动人脸伪造检测技术的发展,特别是在检测视频操纵方面。
DeeperForensics-1.0 is a large-scale face forgery detection dataset co-created by Nanyang Technological University and SenseTime Research. It contains 60,000 videos totaling 17.6 million frames, making it the largest existing dataset of its kind. This dataset constructs a more challenging benchmark with greater scale and diversity by incorporating widespread real-world perturbations. All source videos are meticulously collected, and the forged videos are generated using a novel end-to-end face swapping framework. The quality of the videos generated by this dataset surpasses that of existing datasets, which has been validated via user studies. DeeperForensics-1.0 also includes a hidden test set containing manipulated videos that achieved high deception scores in human evaluation. This dataset is intended to advance the development of face forgery detection technologies, particularly in the area of video manipulation detection.




