UCF101
收藏资源简介:
UCF101是由美国中佛罗里达大学计算机视觉研究中心创建的大型人类动作数据集,包含101种动作类别,共13320个视频片段,总时长约27小时。数据集内容丰富,涵盖了体育、乐器演奏、人类互动等多种场景,视频来源于YouTube,具有真实环境下的摄像头移动和复杂背景。创建过程中,研究人员从YouTube下载视频,手动筛选去除无关内容,确保数据集的质量。UCF101主要用于动作识别研究,旨在解决复杂环境下的动作分类问题,是目前最具挑战性的动作识别数据集之一。
UCF101 is a large-scale human action dataset created by the Computer Vision Research Center of the University of Central Florida, USA. It includes 101 action categories, with a total of 13320 video clips and an overall duration of approximately 27 hours. The dataset has rich content covering various scenarios such as sports, musical instrument performances, human interactions and more. The videos are sourced from YouTube, and exhibit camera movements and complex backgrounds in real-world environments. During its creation, researchers downloaded videos from YouTube and manually filtered out irrelevant content to ensure the dataset's quality. UCF101 is primarily used for action recognition research, aiming to address action classification problems in complex environments, and is currently one of the most challenging action recognition datasets.

- UCF101数据集首次发表,包含101个动作类别的视频数据,成为动作识别领域的重要基准。
- UCF101数据集首次应用于深度学习模型的训练和评估,推动了动作识别技术的发展。
- UCF101数据集被广泛用于各类动作识别算法的比较和性能评估,成为该领域的标准数据集之一。
- UCF101数据集的扩展版本UCF101-24发布,增加了更多的标注信息,提升了数据集的应用价值。
- UCF101数据集在多个国际竞赛和研究项目中被采用,继续推动动作识别领域的创新和进步。



