Wednesday, May 25, 2016

Deep Dream

If you haven't heard of it yet, Google developed this computer vision AI program, Deep Dream (aka: DeepDream, Inception) which tries to interpret the images it's seeing based on images it's already seen. But it gets things wrong. A lot. It's still learning, you know.

Anyways sometime around July 2015, the team figured this could be used to make surreal art by filtering the images as it tried to interpret them:
(Images from Google, @brdskggs/Twitter, & Kyle McDonald/Flickr, via The Telegraph)

It really seems to like dogs, bugs, eyes, arches, cars, and pagodas. And the more times you run a picture, the creepier it gets.
(I'm not sure where this image came from, but it's just as creepy after I had Deep Dream filter it, especially if you click to view the larger version.)

You can try it for yourself, either downloading and setting up the open source software, or just using one of the easy access sites where you upload an image, like this one:

Here's what happened when I ran one of my cosplay picture through once (yeah, I probably haven't mentioned here that I cosplay before, I'm the one on the right):

Time to play spot the difference; the other girl's hand got turned into a dog, there's a ghost above her spear, I seem to have some extra eyes in odd places, the overhang behind me seems to have become a fish, and the girl  in the black dress behind me appears to have grown a demon monkey head, among other things.

You're welcome for the nightmares!

No comments:

Post a Comment

Comments are reviewed for content (to prevent spam), and therefore will not appear right away. Thank you for understanding.