posted on Feb, 21 2018 @ 04:32 AM
a reply to: ArMaP
That´s correct in the sense of stored data, but you don´t compare the stored data in terms of hashes or bits.
What you do is play it and then feed the image stream(!) into a recognition algorithm that will do most of the work. It works neatly even on inverted
Some years ago I used my own self written toolbox to find a small snipped image an ATS member took from a gigapan picture but forgot the location. I
was able to find it pretty fast if you consider that you can zoom in and every zoom layer is a whole new image to crawl over.
Comically, I am unable to find the post from back then, to give you an example. I remember it only took minutes and I was running it on my workstation
inside VS2010, that was running in a VM.
So in the light of that knowledge, I certainly think they(FB) are capable of this in almost realtime on video after another. It´s just a matter of
scaleing it up.
Edit: As a proof, this is the image I used to pinpoint:
maybe it´s possible to track down the thread with this.
To show this is really not a big thing nowadays.
Edit2: Here it is:
edit on 21-2-2018 by verschickter because: (no reason given)