Once one of my non-geologist friends uttered the words “trippy” while staring at a geologic map of North America that was up in the hallway. I agree. To the uninitiated a geologic map probably looks like a fairly random assortment of off-colors, blobs, blebs, sharp lines, and generally strange shapes. As geologists we are trained to take the seemingly random slapping together of colors and lines, and pull from it a history, a piece of the story of our planet. Some engineers at Google challenged a deep learning network with a series of images which produced some “trippy” results. Wouldn’t it be fun to apply their code to some geologic maps?
Basically the program takes an image and attempts to match a some known object (birds, dogs, etc.) in it. Do this for a few iterations and the object becomes more pronounced. For instance they fed it a picture of clouds:
Within lies madness:
Pretty cool, but let’s give it some geologic maps.
When the Trippy filter is applied to the Crooked River map there’s some interesting results. The corners are largely untouched by the dream. Perhaps the machine needs something to ground it to reality?
If your interested, someone turned the deep dream code into a webapp so you can easily upload and make your own dream paintings. If you want to do a bit more digging, the source code is available on GitHub. All of the geologic maps were downloaded or captured from USGS MapView.