The Intelligence Advancement Research Agency (IARPA) has leaked information about the Finder project. A special search engine whose task is to analyze an infinite number of user photos and determine the geographical coordinates of the place where they were taken. Of course, with the good purpose of quickly detecting enemies of the state who love selfies.
The analysis of the geotags of user publications as a whole was adopted by the special services a long time ago, but more or less technically competent people have learned to turn them off in the settings of their gadgets. Therefore, the new concept is to calculate the location of the author of the image using indirect data, and then assign him the same geocode, but only in the internal spy system, for official use.
The predecessor to Finder was a neural network from Google known as PlaNet. With its help, 126 million photographs were processed in open areas with and without geotags in order to train the neural network to recognize the landscape in the background in the frame. The results were not bad, but scanty - if in the same Eastern Ghouta in Syria there was little photographing in general, then there was essentially nothing to work with the neural network.
Therefore, the new goal of the Finder project is to learn how to use data from any sources in order to accurately determine the place on planet Earth where a particular picture was taken. The contractors include Applied Research Associates, BAE Systems, Leidos, and the search engine will be compatible with Aladdin Video and Deep Intermodal Video Analytics tools. The first analyzes the metadata of videos from social networks, the second recognizes signs of "dangerous actions" of people on the video and can use many CCTV cameras at the same time.