AIGC视频取证数据集
收藏资源简介:
本数据集由四川大学网络空间安全学院创建,旨在为AI生成视频的检测提供基准。数据集包含1000个自然视频样本及其对应的1000个AI生成的负面样本,每个负面样本包含24帧。数据集利用多种先进的基于扩散的视频生成算法创建,确保了空间和时间内容的多样性。此外,还考虑了网络传输中的典型视频有损操作,以生成质量下降的样本,用于评估鲁棒性。该数据集主要应用于视频取证领域,旨在解决AI生成视频的检测问题,为未来研究提供基线和挑战。
This dataset was constructed by the College of Cybersecurity, Sichuan University, to serve as a benchmark for AI-generated video detection. It comprises 1000 natural video samples and 1000 matching AI-generated negative samples, each containing 24 frames. The dataset is generated using a variety of advanced diffusion-based video generation algorithms, which guarantees diversity in both spatial and temporal content. Furthermore, typical video degradation operations occurring during network transmission are incorporated to produce degraded-quality samples for robustness assessment. This dataset is primarily utilized in the field of video forensics, with the objective of addressing the challenge of AI-generated video detection and offering a reliable baseline and challenge set for subsequent research.




