search

mingyuan-zhang / MotionDiffuse

MotionDiffuse: Text-Driven Human Motion Generation with Diffusion Model
https://mingyuan-zhang.github.io/projects/MotionDiffuse.html
Other
850 stars 74 forks source link

Question on attention mask type #4

Open makeme-zgz opened 2 years ago

makeme-zgz commented 2 years ago

Thanks for the paper and code! I'm curious on how causal masking and bidirectional masking perform differently in motion diffuse.

mingyuan-zhang commented 2 years ago

Thanks for the paper and code! I'm curious on how causal masking and bidirectional masking perform differently in motion diffuse.

Here we use mask only to eliminate the influence of zero padding. As for valid frames, we consider relationships between all pairs.