The special effects team on Ex-Machina overlaid graphics onto the actor. There were scenes where the actor was "motionless" but they observed that the emotional experience of the film required animated overlaid motion to match the actor.

http://www.theverge.com/2015/5/8/8572317/ex-machina-movie-vi...

"We also put a lot of effort into things like the muscles contracting properly, and the various pipes and wiring having just a tiny amount of jiggle. And it’s something that you really do not notice. But I remember when we were looking at shots, for whatever reason when we put a shot through to render overnight that secondary animation hadn’t rendered properly, so it was missing. And everything suddenly felt very stiff. And you kick the shot off again, this time with that animation integrated into it, and it works again. It’s not something that you can necessarily put your finger on as being wrong, but if it’s missing then you suddenly feel that something’s strange."

Here's another video of the algorithm being tested against a wider range of targets.

https://www.youtube.com/watch?v=AGA6eJXZAOw

The ability to pick up saccades could have interesting HCI implications if the signal is consistent.

It would be interesting (and probably very funny) to apply this to a video of a person trying to stand as still as possible.

I remember a similar technique (maybe it's from the same people) where they amplified color changes.

So the tiny flushes caused by the heartbeat were clearly visible

https://www.youtube.com/watch?v=ONZcjs1Pjmk (from 1m30s)

Yup, Durand and Freeman worked on both. The Eulerian video magnification isn't just limited to color changes, of course, as the video shows. This new technique is still based on Eulerian magnification, if I understand correctly.

And the Person can see itself in a "mirror" that shows the magnified movements, thus making drastic adjustments and shortly losing all control.

similar to this? https://www.youtube.com/watch?v=lQAXtTpm930 (based on the technique raverbashing linked)

Interesting talk by one of the researchers on the techniques including recovery of audio from video of passive object vibrations:

http://www.ted.com/talks/abe_davis_new_video_technology_that...

If you realize that a camera is a time-series motion capture sensor, as is a microphone (the position of the diaphragm) then this is actually less relevatory...

Honest questions: how many "frames per second" is a standard microphone? What exactly is a "frame" of audio?

44.1 kHz (CD quality audio) is 44,100 samples per second, though professional recording equipment typically supports up to 96 kHz sample rates.

Note that for the more common analog microphones, the microphone itself doesn't determine the sample rates; it's the digital sampler to which it's connected.

"the microphone itself doesn't determine the sample rates"

True, but it has a response curve, and I assume most don't respond as well to >20kHz frequency (or even less)

This makes sense, thank you!

Keep in mind that, even with 120 fps of video, you can pretty much only reproduce frequencies of up to 60 Hz. I guess they do some more modeling to make the sound more intelligible, though.

Typical sampling rates for microphones are 44.1, 48, or 96 kHz. A "frame" of audio would just be a sample taken from the microphone

That motion was not Imperceptible.

The video that @raverbashing mentined seams to be much more useful.

That's an older technique that this is based on -- the one that, as mentioned, can't handle both large and small motions. Given that, I wouldn't compare it by that one example.

How does it compare to Eulerian Video Magnification?

http://newsoffice.mit.edu/2012/amplifying-invisible-video-06...

That's the older technique that this is based on. See the actual paper: http://www.cv-foundation.org/openaccess/content_cvpr_2015/pa...

I can imagine this being useful to show how buildings and bridges move in the wind or earthquakes. Could be quite alarming though.

What is this sort of thing typically used for?