Thanks for your great work.
As in the paper, b_axis is computed by (child_joint - current_joint),
But in axislayer.py, transf[..., :3, :3](which seems to be global rot?) mul b_axis after that. Is it for translating b_axis local to global? As axis-angle in local, why trans b_axis to global. I tried remove this operation and found the axis weired(b_axis not direct its child).
In skinlayer, output joints isn't same as mano joints. So how to use the fit result in usual MANO.
I'm confused as a new learner. Hope for your reply!
Thanks for your great work. As in the paper, b_axis is computed by (child_joint - current_joint), But in axislayer.py, transf[..., :3, :3](which seems to be global rot?) mul b_axis after that. Is it for translating b_axis local to global? As axis-angle in local, why trans b_axis to global. I tried remove this operation and found the axis weired(b_axis not direct its child).
In skinlayer, output joints isn't same as mano joints. So how to use the fit result in usual MANO. I'm confused as a new learner. Hope for your reply!