VIDS-Guard Dataset (v1.0, Part 2 of 2)

Part 2: VIDS-Guard Dataset (v1.0, External Test Set – Part 2 of 2) This record contains the rest of VIDS-Guard main dataset as well as the External Test Set (v1.0), created to evaluate the cross-dataset generalization capability of deepfake video detection models. The test set includes 600 unseen videos (300 real / 300 fake) selected from independent sources such as HiDF (2025), CelebV-HQ unseen subset, and several Kaggle datasets. None of these samples appear in the main VIDS-Guard training set. Videos are provided in .mp4 format under /ExternalTest/Real and /ExternalTest/Fake directories, ensuring balanced evaluation across new manipulation techniques and authentic clips. This dataset is intended for evaluation only and should not be used for model training. Access is restricted to verified academic researchers for non-commercial use. Note: This record is Part 2 of 2. Please also refer to VIDS-Guard Dataset (v1.0, Part 1) under DOI [10.5281/zenodo.17362749] for the main dataset and full documentation. Full dataset details and references are provided in the attached README file.第二部分：VIDS-Guard数据集（v1.0，外部测试集——共两部分之第二部分） 本数据集包含VIDS-Guard主数据集的剩余部分，以及旨在评估深度伪造视频检测模型（deepfake video detection models）跨数据集泛化能力的外部测试集（v1.0）。 该测试集包含600条未出现在主训练集中的视频（300条真实视频、300条伪造视频），其来源涵盖HiDF（2025）、CelebV-HQ未见过样本子集以及多个Kaggle数据集等独立数据源。所有上述样本均未在VIDS-Guard主训练集中出现。 视频以.mp4格式存储于/ExternalTest/Real与/ExternalTest/Fake目录下，可确保对新型篡改技术与真实片段开展均衡评估。本数据集仅用于模型评估，不得用于模型训练。 本数据集仅限已认证的学术研究人员用于非商业用途。 注：本记录为共两部分的第二部分。如需获取主数据集与完整文档，请参阅DOI为[10.5281/zenodo.17362749]的《VIDS-Guard数据集（v1.0，第一部分）》。 完整数据集详情与参考文献详见附件README文件。