Some questions about Style transfer

[JimHeo]

Hi, thanks for your great work.

I have some questions about your style transfer's.
Before the question, I ask for your understanding because I am not familiar with this animation domain yet.

I would like to ask the following questions:

1. In convert_to_disc() function, I am wondering the reason about using the velocity(especially local velocity) for the input of discriminator.

2. Also, before convert_to_disc() function, merge_pos_glb() function is used for merging joint position, global root position and global root rotation. Is the function(and also global variable) essential?
   Because of the mention in the paper, "Note that the global root positions are discarded from the representation of the networks's input/output,", I am wondering the role of the merge_pos_glb() function.

3. I don't understand that the DoF of global rotation is 1. Could it be recovered to euler or quaternion?

4. Could I use different dataset for each Content and Style? (i.e. SMPL for Content, CMU for Style, although different number of joints)

5. Could I use rotation for Style with trimming the number of joints?

Thanks.

Best Regards,
Jim.