[CVPR2025]HumanDreamer: Generating Controllable Human-Motion Videos via Decoupled Generation
MotionVid Dataset
MotionVid is a dataset containing 1.2M text-pose-video pairs. The videos are sourced from the internet and public datasets, the poses are extracted using DWPose, and the text descriptions are generated using ShareGPT4Video.
Note: The repository does not include video files due to licensing restrictions. If you require the video files, you must download them… See the full description on the dataset page: https://huggingface.co/datasets/chuanshuogushi/MotionVid.
