Motion-X++
收藏资源简介:
Motion-X++是由国际数字经济学院(IDEA)等机构开发的大规模多模态3D全身人体运动数据集,旨在解决现有数据集在面部表情、手势和精细姿势描述方面的不足。该数据集包含19.5M帧的3D全身姿势注释,覆盖120.5K运动序列,数据来源于80.8K RGB视频和45.3K音频。数据集通过自动化的注释流程生成,支持多种下游任务,如文本驱动的全身运动生成、音频驱动的运动生成、3D全身人体网格恢复和2D全身关键点估计等。Motion-X++的应用领域广泛,涵盖了机器人、动画、游戏和生成艺术等多个领域,旨在生成更具表现力和自然感的运动序列。
Motion-X++ is a large-scale multimodal 3D full-body human motion dataset developed by the International Digital Economy Academy (IDEA) and other institutions. It aims to address the shortcomings of existing datasets in describing facial expressions, gestures and fine-grained poses. This dataset contains 19.5 million frames of 3D full-body pose annotations, covering 120.5K motion sequences, with data sourced from 80.8K RGB videos and 45.3K audio clips. Generated via an automated annotation workflow, it supports a wide range of downstream tasks, including text-driven full-body motion generation, audio-driven motion generation, 3D full-body human mesh recovery, and 2D full-body key point estimation, among others. Motion-X++ has broad application scenarios across fields such as robotics, animation, gaming and generative art, and is designed to generate more expressive and natural motion sequences.




