
Hieronymus Bosch’s “The Garden of Earthly Delights” (Image via Kyle McDonald/Flickr) (click to enlarge)
You may have thought Hieronymus Bosch’s “The Garden of Earthly Delights” could not get any more trippy, but researchers have developed new software that proves that yes, the Netherlandish painter’s baffling triptych just needs a little bit of code to become a complete mind-bender. Last month, a team of Google software engineers wrote about a new visualization tool to more closely examine the structures of artificial neural networks. By dissecting these networks, the team hoped to better understand how they learn, layer-by-layer, to generate the true essence of whatever they are replicating.
After feeding an image to a network for it to analyze, the researchers then “pick a layer and ask the network to enhance whatever it detected.” As they explain:
Each layer of the network deals with features at a different level of abstraction, so the complexity of features we generate depends on which layer we choose to enhance. For example, lower layers tend to produce strokes or simple ornament-like patterns, because those layers are sensitive to basic features such as edges and their orientations.

Joseph Ducreux’s self-portrait (Image via @thebyronichero/Instagram)
These networks, therefore, can even over-interpret pictures — and may be applied to any image — to morph parts of reality into other bits of reality, so the dream-like, dizzying results are oversaturated with snippets of other images. While this is useful for their own work, the researchers also wondered “whether neural networks could become a tool for artists — a new way to remix visual concepts — or perhaps even shed a little light on the roots of the creative process in general.”
So Google open sourced the code and invited the public to generate their own works of art, encouraging people to tag their creations with #deepdream. The results are predictably varied, but many have chosen classic works from art history as their targets, transforming familiar pictures into baffling but often comical ones. Because the original network was trained to identify mostly images of animals, plenty of dogs, birds, slugs, and fish pop up in places where they don’t belong. Claes Oldenburg’s sculpture “Pastry Case” is slightly less appetizing when the baked goods it contains appear to be sprouting slimy creatures; one can’t help but chuckle at Joseph Ducreux, (already famous as a meme) when his left hand is a tiny sloth’s face.
Other furries show up in Edvard Munch’s “The Scream,” in which a yeti covers its ears while a basset hound peers out from a sea of eyeballs; or in fellow Norwegian painter Lars Hertervig’s “Skogtjern (Forest Lake),” which is completely overrun by “mutant dogs.” The deep dream code produces the most amusing results when it’s applied to austere works of art, but be warned when perusing the hashtag: it does have the power to ruin the quotidian.

Applying “deepdream” to Claes Oldenburg’s “Pastry Case” (Image via Kyle McDonald/Flickr)

Processing George Seurat’s “A Sunday Afternoon on the Island of La Grande Jatte” (Image by Matthew McNaughton via Michael Tyka/Google) (click to enlarge)

Michelangelo’s “Creation of Adam” (Image via Kyle McDonald/Flickr)

Another take on Hieronymus Bosch’s “The Garden of Earthly Delights” (Image via @aut0mata /Twitter) (click to enlarge)

Edvard Munch’s “The Scream” (Image via Michael Tyka/Google)

Lars Hertervig’s “Skogtjern (Forest Lake)” (Image via @matsiyatzy/Twitter)

Leonardo da Vinci’s “La Joconde” (Image via Dimiter Stanev/Facebook)

Salvador Dali’s “The Persistence of Memory” (Image via Kyle McDonald/Flickr)

Sandro Botticelli’s “The Birth of Venus” (Image via Kyle McDonald/Flickr)
Disturbed. Now watch Fear & Loathing in Las Vegas run through Google’s trippy deep dream software https://youtu.be/oyxSerkkP4o
Wow, how to make great art works less interesting. What an accomplishment.