I work at an ML company, and in the corporate slack someone posted a iPhone photo of a freeway and some mountains.
I zoomed in, found it on Google Maps, and posted "It's here (https://goo.gl/maps/s4mMkeiZWJp9WZSY9). Beautiful mountains."
Response: "How did you do that?"
Me: "Clues are in the photo, though blurry. Freeway and Utah state route outline means probably I-15 (I-80 might also have exit numbers that high, but it would be out in the country). Exit numbers and their sequence fixes a location and direction."
Response: "Bah - I was hoping someone had created a tool that could recognized where a pic was taken by x-ref to Google Streetview. If that doesn't exist, some smart person should make it."
OP: "there is a training set for location by image: http://places.csail.mit.edu/"
There's also a demo, where you can upload your own photo and have their CNN (convolutional neural network) analyze it: http://places2.csail.mit.edu/demo.html
Example:
(https://i.imgur.com/rff8szc.png)