Quote from: whiterabbit on Jul 07, 2015, 11:36:20 PM
I don't get it. It knows a 1000 words and is using fractal mathematics to enhance edges into objects?
The way it was described to me was Googles image recognition software would scan the picture and find anything that looked even remotely similar to anything in it's database (this is using it's own set of sources though, not what Google image recognition would normally use). It will make minute adjustments to make the picture look slightly closer to what it recognize, and then look at it again. Here's an example.
Take this picture.It is a cat laying down. Now, have Google run it through it's recognition loop.
NO ONE TOLD IT TO MAKE IT DOGS. That's just what it saw.
Sadly, we haven't been able to see the process in real time. We can watch the Twitch stream, but we can't add our own photo to the process and see what Google does.
Note, I found out about this while walking in from seeing Genisys in the cinemas. Was kind of freaky.