Google knows about us more than we know about ourselves. It’s a little bit creepy but it’s true. The company had long been tracking our location with the service called “location history” which has been turned into a feature called Google Timeline available in Google Maps app.
Now the company has tools to know even more about our whereabouts. Today, MIT Tech Review reported on a new project called PlaNet led by Tobias Weyand, a computer vision specialist at Google, who have trained a deep-learning machine to guess the location of almost any photo based just on pixels.
The system is based on a Neural Network which has been fed with over 90 million geotagged images across the planet, and trained to spot patterns based on location. The network has learned to detect language on labels and displays, types of vegetation, architectural styles among other parameters.
The first batch of trials proved promising – PlaNet guessed the country of a photo with a 28.4% success rate and the continent – with a 48% success rate. PlaNet was also able to recognize 3.6% of times images at street level accuracy, and 10% at city-level. But Weyand decided to give it the ultimate test using our favourite StreetView game -Geoguessr. “In total, PlaNet won 28 of the 50 rounds with a median localization error of 1131.7 km, while the median human localization error was 2320.75 km,” the author said.
It’s fascinating and scary that our location can be determined based on almost anything. I guess that we live in times that in order to keep your whereabout just for yourself you need to stay offline.