For Facebook and Google, it’s not enough for computers to recognize images… they should create images, too. Both tech firms have just shown off neural networks that automatically generate pictures based on their understanding of what objects look like.Facebook’s approach uses two of these networks to produce tiny thumbnail images. The technique is much like what you’d experience if you learned painting from a harsh (if not especially daring) critic. The first algorithm creates pictures based on a random vector, while the second checks them for realistic objects and rejects the fake-looking shots; over time, you’re left with the most convincing results. The current output is good enough that 40 percent of pictures fooled human viewers, and there’s a chance that they’ll become more realistic with further refinements.
Google’s take heads in the opposite direction. Instead of striving for realism, it’s producing art by letting the neural network run wild and decide on the visual elements that it wants to emphasize. If you give the machine a photo of the sky and it thinks there are birds in the scene, it’ll keep amplifying those avian traits until they’re impossible to miss. The finished work is more than a little trippy, especially if you give it random noise as its source material — as you can see above, the results give impressionist and surrealist painters a run for their money. You’re not likely to see these Facebook and Google programs replacing human artists and photographers, but they’re skilled enough to draw images you might enjoy.
Source: Facebook (ArXiv.org), Google Research Blog
Copied from: Engadget