generative-models icon indicating copy to clipboard operation
generative-models copied to clipboard

Published 20 hours ago • verified publisher icon Stability-AI

Positional encoding for time for temporal attention layers in SVD

Open jeanne-wang opened this issue 1 year ago • 1 comments

Hi, do you use positional encoding for time for temporal attention layers in the SVD model?

jeanne-wang avatar Jan 12 '24 02:01 jeanne-wang

None, but from their code, I think such cues are fed to temporal blocks via frame index embedding.

Little-Podi avatar Jan 17 '24 10:01 Little-Podi