Google will find anywhere on the planet of its new search engine

Posted By on February 26, 2016

Try to download the photo on the internet and find out where it was made, using only the data that can be seen in the picture. Well, if the frame contains the world-famous buildings or famous places like Niagara Falls. But what to do when nothing like the picture is not, and is not enough geotagging in the file properties?

In Google figured out how to simplify the search, here’s the solution to this puzzle. Veyand Tobias (Tobias Weyand) with his friends engaged in the creation of a system that will compare the pixels in the image with their image database.

The search will compare similar images in my memory with the specified file. The database contains 126 million images from geotmetkami learning system with artificial intelligence. The whole planet was divided into 26,000 sites, their size depends on the number of pictures that made in different areas.


So the big city in this case, a better chance of identification than the far corners of the earth, where rarely photographers. In addition, the search engine does not work with expanses of water and cold polar areas, where photographers are extremely rare.

After this, the development team has collected pictures geotagged from the Internet in a huge database. The system operates independently from the 91 million images. Then it was added another 34 million files to check whether it is correctly searched.

This tool is called PlaNet, the first full-fledged test was the work with photos from Flickr. From there, 2.3 million photos uploaded geotagged for testing. That’s what the test showed. So far, the service is not so accurate, he correctly identifies the country of origin of photos of 1 in 3, and the continent on every second shot. It is right in the city only 1 in 10, and the street is 3 times less.

But that is not all. To compare the ability of artificial intelligence and a dozen travelers Tobias Veyand created the game. It turned out that PlaNet won 28 games out of 50 with an average error in 1131 km, while the people were wrong 2320 km.

You too can play it for free and without registration SMS. :) The problem is this: you throw on a certain street, using Google Street View system, and then you are asked to specify your location.

In any case, the system knows and remembers a lot more information than the most active explorer in the world, constantly learning and gradually focused even indoors. The program requires at least 400 MB of RAM, so if the service is adapted for smart phones, will be very good.

Imagine, you get acquainted with the girl, she sends you her selfie on the background of the carpet of his court, you’re driven into the search engine, bam, you find this place! And waiting for her in the evening with a bouquet.