Uncanny Valley Of The Apes

There is a new tool.

Researchers in Germany and Belgium created this.

It's called Mac Action.

It's using deep learning to build frameworks fully animated 3D avatars of monkeys, multi-camera video footage, actual movements, that whole thing.

You can't apparently do standard motion capture on a monkey the way that you'd do it on a human actor in a film.

Why? The researchers said that reflective markers don't work on an animal that isn't going to sit through a fitting.

Okay.

Yeah, look, that's fair.

Honestly, I assumed they just sedate the monkeys.

So I think it's kind of nice that they're treating them this well.

That's heartening to me.

So to do this, they had to develop an entire new markerless tracking method that was accurate enough for computer animation.

That's most of the actual engineering challenge with this whole thing.

And then once they had the avatar...

they needed to know if it looked convincing to a monkey, not just to a human.

So they showed eight male Reese's monkeys, two sets of videos.

Hot singles in the area.

Pretty much.

Real footage of monkeys and the animated avatar that was performing identical movements.

And then they measured where the monkeys looked and how often and how long.

on each one.

Apparently the viewing patterns were indistinguishable between the real video and the animation.

So they had the same attention, the same duration, same points of interest on the body.

So functionally, when you looked at it, the monkeys were treating the avatar like it was a real monkey.