UCF101-AD
收藏资源简介:
UCF101-AD是一个大规模的动作否定基准数据集,旨在评估视频多模态大语言模型在存在误导性上下文和运动线索的情况下,能否正确拒绝视频中实际不存在的动作。该数据集从公开的YouTube视频中提取了短视频片段,专门用于研究、教育和非商业目的。根据其声明,该数据集的使用符合美国版权法第107条关于合理使用的规定,因其具有转换性、仅使用有限部分、非商业意图且服务于公共利益,不会替代原始作品。数据集不主张对原始内容的所有权,版权仍归原持有者所有。数据集以“原样”形式提供,不附带任何担保或责任。使用者需遵守隐私和伦理准则,避免用于有害用途,若认为权利受到侵犯可联系发起方提出下架请求。
UCF101-AD is a large-scale action rejection benchmark dataset designed to evaluate whether video multimodal large language models can correctly reject actions that do not actually exist in a video when faced with misleading contextual and motion cues. This dataset extracts short video clips from publicly available YouTube videos, and is exclusively intended for research, educational, and non-commercial purposes. Per its statement, the use of this dataset complies with the fair use provisions of Section 107 of the United States Copyright Act, as it is transformative, uses only a limited portion of the original content, has non-commercial intent, serves the public interest, and does not substitute the original works. The dataset does not claim ownership of the original content, and all copyrights remain with their original holders. The dataset is provided "as-is", without any warranties or liabilities. Users must comply with privacy and ethical guidelines, refrain from using it for harmful purposes, and may contact the initiating party to request removal if they believe their rights have been infringed upon.





