HiC-Motion
收藏资源简介:
HiC-Motion是一个大规模的人类运动视频数据集,由Adobe研究所和加州大学默塞德分校的研究团队创建。该数据集包含30万条视频,涵盖了室内外多种场景和多样化的日常及体育活动。数据集通过使用先进的3D姿态估计算法对视频进行标注,确保了运动序列的高质量和一致性。创建过程中,研究人员从开放域的互联网视频中筛选出包含单人运动的视频,并进行了详细的预处理和增强。该数据集主要用于2D场景下的人类运动生成任务,旨在解决现有方法在特定运动类型和场景匹配上的局限性,提升视频合成中的人类运动质量。
HiC-Motion is a large-scale human motion video dataset developed by research teams from Adobe Research and the University of California, Merced. This dataset comprises 300,000 videos, covering diverse indoor and outdoor scenarios as well as a wide range of daily and athletic activities. All videos in the dataset are annotated with advanced 3D pose estimation algorithms, ensuring high quality and consistency of the motion sequences. During the dataset construction process, researchers screened videos containing single-person human motions from open-domain internet videos, followed by detailed preprocessing and data augmentation. This dataset is primarily designed for human motion generation tasks in 2D scenarios, aiming to address the limitations of existing methods in matching specific motion types and scenarios, and to improve the quality of human motion in video synthesis.




