ml-hugs
ml-hugs copied to clipboard
If the smpl params update?
Thanks for your great work, in human avatar papers, most of them will update smpl params (betas, body_pose, etc.) during training as the estimated one not exactly accurate, when I go through your code, it seems that these params not update, instead given more accuracy estimated smpl params for input.
Could you please share the benefits of this approach? Or my understand is not correct?