It’s really great to see nvidia making such a useful tool freely available in such an easy to use way. Other video based pose tracking tools are either poor or cost money.
I’ve tested it with a few videos so far and it can even work somewhat with sports and more active movements. Though it does lose tracking quite a bit you can really see the potential when it does work.
It seems to handle standing stuff pretty well but sadly falls apart pretty quickly if the person moves in other awkward ways like with various sports, exercises, acrobatics etc. It actually worked pretty well in some situations and knew if a person was facing the camera or not which was surprising.
Im not sure how much ai and physics predictions could add but i think nvidia may have some research going on already in this area and hope they will make some big improvements. It’s possible to get pretty amazing tracking i believe if done right. Im guessing this is an early version of what they’re working on. I hope they make it a priority as i think good and easy animation will be the one of most useful features for machinima.