Ever seen at a photo of a delicious looking meal on Instagram and wondered how the heck to make that? Now there’s an AI for that, kind of.
Facebook’s AI research lab has been developing a system that can analyze a photo of food and then create a recipe. So, is Facebook trying to take on all the food bloggers of the world now too?
Well, not exactly, the AI is part of an ongoing effort to teach AI how to see and then understand the visual world. Food is just a fun and challenging training exercise. They have been referring to it as “inverse cooking.”
According to Facebook, “The “inverse cooking” system uses computer vision, technology that extracts information from digital images and videos to give computers a high level of understanding of the visual world,”
The concept of computer vision isn’t new. Computer vision is the guiding force behind mobile apps that can identify something just by snapping a picture. If you’ve ever taken a photo of your credit card on an app instead of typing out all the numbers, then you’ve seen computer vision in action.
Facebook researchers insist that this is no ordinary computer vision because their system uses two networks to arrive at the solution, therefore increasing accuracy. According to Facebook research scientist Michal Drozdzal, the system works by dividing the problem into two parts. A neutral network works to identify ingredients that are visible in the image, while the second network pulls a recipe from a kind of database.
These two networks have been the key to researcher’s success with more complicated dishes where you can’t necessarily see every ingredient. Of course, the tech team hasn’t stepped foot in the kitchen yet, so the jury is still out.
This sounds neat and all, but why should you care if the computer is learning how to cook?
Research projects like this one carry AI technology a long way. As the AI gets smarter and expands its limits, researchers are able to conceptualize new ways to put the technology to use in our everyday lives. For now, AI like this is saving you the trouble of typing out your entire credit card number, but someday it could analyze images on a much grander scale.