how to use in-the-wild video data to train the generative model

[Crayon-Shinchan]

How to use in-the-wild video data to train the generative model? I am confused about how canonical human images are obtained in the pre-training stage and how multi-view images are obtained.

[lingtengqiu]

we pretrained rgb-normal diffusion model for animation human like CHAMP. Then we fine-tune multi-view rgb-n diffusion model.