 | ||
| Sleeping Beauty has awakened. (source) GoogleNet Places205, inception_4b/pool_proj, 10 iterations, 10 octaves, scale 1.2 |
Google’s DeepDream has been called “dazzling, druggy, and creepy” (Slate), “weirdly mesmering” (Wired UK), and “utterly psychedelic and surreal […] Lovecraftian nightmare fuel” (Le Bonbon). One month after Google Research loosed their creation on the world, this artificially intelligent image recognition-turned-generation neural net has turned vast swaths of the internet into dark, globulous, glistening nests of hairy eyes and alien dog-worm larvae. Techies and laymen alike have now used DeepDream to profane everything from pizza and presidents to, unsurprisingly, porn. (Also, ahem, brass instruments—see previous post.)
But even AI needs a break from puppyslugs. What else can it do? Well, as of mid-July, various incarnations of DeepDream have been specifically trained on images of flowers (8,000), faces (27,000), cars (137,000), and, oh, ~2.5 million images of PLACES.
Bingo. DeepDream’s new job: ARCHITECT.