ali-vilab / UniAnimate

Code for Paper "UniAnimate: Taming Unified Video Diffusion Models for Consistent Human Image Animation".
https://unianimate.github.io/
1.03k stars 60 forks source link

噪声被直接cat到pose上,是否会导致信息泄露? #41

Open xfguo-ucas opened 4 months ago

wangxiang1230 commented 4 months ago

Thank you for your attention. In fact, in the network, we want to inject post information into the 3d-unet, which is more beneficial for us to generate pose-driven videos.