Animate Anyone: Consistent and Controllable Image-to-Video Synthesis for Character Animation
14.23k
stars
952
forks
source link
为什么要采用Reference Image的信息用Spatial-Attention和Cross-Attention编码到Pose Sequence的Denoising UNet里面的方法而不是反过来 #46
Open
hxypqr opened 8 months ago
有什么数学解释说明采用Reference Image的信息用Spatial-Attention和Cross-Attention编码到Pose Sequence的Denoising UNet里面的方法比反过来更有优势吗