GenVidBench:A Challenging Benchmark for Detecting AI-Generated Video
Zhen-Liang Ni*, Qiangyu Yan*, Tianning Yuan, Mouxiao Huang, Yehui Tang
Hailin Hu✉️, Xinghao Chen✉️, Yunhe Wang✉️
▶ Huawei Noah's Ark Lab
Instruction
The rapid advancement of video generation models has made it increasingly challenging to distinguish AI-generated videos from real ones. This issue underscores the urgent need for effective AI-generated video detectors to prevent the dissemination of false information through such videos. However, the development of high-performance generative video detectors is currently impeded by the lack of large-scale, high-quality datasets specifically designed for generative video detection. To this end, we introduce GenVidBench, a challenging AI-generated video detection dataset with several key advantages:
- Cross Source and Cross Generator: Mitigates the interference of video content on the detection. The cross-generator ensures diversity in video attributes between the training and test sets.
- State-of-the-Art Video Generators: Includes videos from 8 state-of-the-art AI video generators.
- Rich Semantics: Videos are analyzed from multiple dimensions and categorized by content.
We conduct a comprehensive evaluation of different advanced video generators and present a challenging setting.
News
2025/02/11: Update the download link and fixed the bug. Note that we uploaded the ID of Pair1 in the original dataset to the file path 'data/sampled_dataset_uuid.zip'. Due to copyright restrictions, We can't provide these videos directly. You can select the corresponding video from these three data sets Vript/HD-VG-130M/VidProM based on these IDs.
2025/01/25: The training code is released.
2025/01/20: The paper is published at Arxiv.
Reference
✨ If you find our data helpful, please cite our paper or point a star. Thank you very much!
@misc{ni2025genvidbenchchallengingbenchmarkdetecting,
title={GenVidBench: A Challenging Benchmark for Detecting AI-Generated Video},
author={Zhenliang Ni and Qiangyu Yan and Mouxiao Huang and Tianning Yuan and Yehui Tang and Hailin Hu and Xinghao Chen and Yunhe Wang},
year={2025},
eprint={2501.11340},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2501.11340},
}
License
CC BY-NC 4.0