It probably won't be long before algorithms can mocap just from the recorded footage and then transpose those movements to arbitrary 3D models :/

That doesn't take away from the beauty of martial arts though. You can have an exact definition of all the moves you'll ever do, but still spend decades refining them.

For actors though... yeah I can see why they're worried

It's already very possible with off-the-shelf OpenPose. I can only imagine that the proprietary systems are more powerful and have better performance.

Good to know. Thanks for sharing!

I’ve seen the new Indiana Jones last week, and found the (clearly) animated opening scenes quite unrealistic. There was constantly something wrong with Indy’s sudden movements and other tells. Felt more like a really good cutscene. And they did have mocap available for that. I don’t think you‘ll get believable virtual actors (outside of action scenes) without mocap anytime soon.

The Avatar films, the latest Planet of the Apes, and some of the Gollum scenes from the LOTR films seem like peak mocap. I'm guessing it's partly do to the actors, the director, the digital artists, and the fact they aren't all truly human. Things not being human make it easier to fool our eyes because we look at humans every day since birth.

I would love to hear from a primatologist who's decades into their career to see if they think Planet of the Apes holds up.

It's definitely rudimentary possible.

During covid I was working on a paintball video game and I took pro paintball footage and ran some university research skeleton modeling program on it and got reasonably good results. I was going to use that for crouch and bunker hugging animations. And that was just me an HN reader playing around.

EDIT: Sibling posted and I was using that project -- OpenPose!

They can do that today