CoMoVi: Co-Generation of 3D Human Motions
and Realistic Videos

Chengfeng Zhao¹, Jiazhi Shu², Yubo Zhao¹, Tianyu Huang³, Jiahao Lu¹,
Zekai Gu¹, Chengwei Ren¹, Zhiyang Dou⁴, Qing Shuai⁵, Yuan Liu^{1 †}

¹HKUST    ²SCUT    ³CUHK    ⁴MIT    ⁵ZJU   
^{† Corresponding author}

📝 TODO List

• Release inference and training code
• Release data preparation pipeline
• Release CoMoVi Dataset

Acknowledgments

Thanks to the following work that we refer to and benefit from:

• VideoX-Fun: the video generation model training framework;
• CameraHMR: the excellent SMPL estimation for pseudo labels;
• Champ: the data processing pipeline