Overview of our method. We reformulate the 4D video creation as video inpainting task. Left: given a video, we first generate the composite masks from the dynamic point cloud and feed them into video inpainting model to unlock its 4D video creation capability. Right: To unlock the capability of generating 4D video with larger motion, we first generate videos with small motion, then feed them into the model to improve the temporal consistency progressively.
💖Enjoyed this project? Drop a star ⭐follow-your-creation⭐—it means a lot!
🚀Our project page is borrowed from ISA4D.💡
@article{
title={Follow-Your-Creation: Empowering 4D Creation through Video Inpainting},
author=
journal={},
year={},
publisher={}
}