BEACON TRANSCRIPT – Google’s working on a new search engine that could make geotags a thing of the past. The new tag engine is called PlaNet and it is able to tell you where the picture was taken with an uncanny accuracy. As described by its makers, PlaNet is the embodiment of globetrotting, because it has seen far more places than any human being.
Presently, Google is working on a way to improve geotagging. The result of their research is an innovative search and ID engine called PlaNet. What does planet do? Well, it can tell you where a picture was taking without the need of looking at a geotag.
Weyland, the mastermind behind Google’s brainchild declared that the engine is capable of recognizing so many places by looking at squares. As we know, the squares in a picture have different sizes, and the more pictures taken in a specific location, the more squares there are. PlaNet uses these squares in order to pinpoint the exact location of the picture.
Furthermore, the search and ID engine will far even better in a crowded city like Paris or London than in remote locations. For example, if a tourist takes a picture of the Eiffel Tower, the engine will recognize the dimensions of the squares associated with the Parisian landmark and will say that the picture was taken in Paris.
How was this feat achieved? In order to hone the geo-recognition technology, Weyland and his team of computer researchers uploaded approximately 91 million pictures into the software’s databanks. Surprisingly enough, the computer did not try to commit to memory all the pictures it has been fed, but rather tried to remember the size of each pixel associated with a certain picture.
In order to take it to the next level, the team devised a neural network, which allowed the system to recognize pictures from different areas of the globe. All in all, the team uploaded no less than 126 million pictures into the system.
After the computer processed the first 91 million pictures, it devoured the remaining 31 million. What was left to do? To put the software to the test of course. To test out the system, the team used a group of 2.3 million pictures.
According to Weyland’s statement, PlaNet was able to recognize the location depicted in the pictures with 28.4 percent accuracy. Moreover, it would seem that this number rose to 48 percent when the computer was asked to recognize the continent.
Even though the numbers seem pretty low, then you should know that PlaNet was off by 1137.7 km, whereas a human asked to recognize those exact location was off by 2320.75 kilometers.
Indeed, PlaNet is the embodiment of globetrotting, making every wanderer look like someone who just discovered that the world isn’t flat. With 28 out of 50 won rounds, the software is sure to replace geotagging in the nearby future.