I gave it my best shot. I'm sure another iteration or two would make it better. Dataset of 71, superhero comics, and mostly Superman at that.
I tried to teach it the difference between arm and hand poses.
Arm poses: Arm_up, arm_down (can be combined) arm_back (like flexing with the other arm up) arms_up, arms_down, arms_behind, arms_out
hand poses: clenched_hand, clenched_hands, flat_hands, open_hands
Oh and legs_together of course!
I hope you like it :)
Description
Just a quick transfer of the Illustrious dataset. It does a little better I think, but probably still isn't consistent