The problem with searching for images you or your friends have posted on Facebook is that it is kind of hard to find them unless they are properly tagged or you know when they were taken. Facebook is trying to solve this problem with the recently announced AI image search system that is capable of seeing and understanding the content of a photo.
For example, if you’re looking for a photo that isn’t tagged but you remember you were wearing a black shirt when it was taken, simply search for “black shirt photo”. The system will then show you all the images containing a black shirt and therefore save you a lot of time you would have otherwise wasted trying to find the right image yourself.
In addition to clothing items such as a shirt or a scarf, the system is also capable of detecting scenes (garden), objects (bicycle), animals (dog), as well as places and attractions (the Golden Gate Bridge).
The new AI system uses the Lumos platform that the company built specifically for image and video understanding. In order to make sure that the search results are relevant to the query, it has to have a good understanding of the photo’s content. Facebook has accomplished this by training the system to detect different objects by processing billions of different images.