search

ali-vilab / UniAnimate

Code for Paper "UniAnimate: Taming Unified Video Diffusion Models for Consistent Human Image Animation".
https://unianimate.github.io/
982 stars 52 forks source link

Why your algorithm can generate better hands #36

Open xiaohutongxue-sunny opened 3 months ago

xiaohutongxue-sunny commented 3 months ago

Bro, why your work can generate a nice hands, it's effective because of your noval algorithm or your sufficient data? If the data is only 2000 , can it work well too?

wangxiang1230 commented 3 months ago

Hi, you can refer to the paper for more details. In short, the advantages of our method are as follows:

  1. Unified 3D-Unet model is adopted to facilitate appearance alignment;
  2. Introduce ref pose to guide learning corresponding source-target pose relationship.