Google wants to guess meal calories using food pics

If you've an Instagram account (or a Twitter account, or a Facebook account, etc.) you've likely found yourself snapping a picture of a meal too good to devour without first having recorded its existence. Those food pictures dot the digital landscape, and Google has big plans for them: calorie counting. In the future, or at least the future where this Google product both exists and works well, folks will be able to monitor their daily calorie counts simply by snapping a picture of their food before chowing down.

The project was recently detailed at the Rework Deep Learning Summit, where Google researcher Kevin Murphy detailed an effort to use deep learning algorithms for estimating a meal's calories based on a picture of it. The project has been dubbed "Im2Calories", and it seems to work — at least presently — by identifying the food items in the image and approximately how large each one is.

The project is able to analyze and estimate food items found in ordinary snapshots, such as the ones you'd find on Instagram, rather than just higher-resolution photographs. It likewise takes into account things like condiments, which increase the overall calorie numbers. All this could be used for something as simple — and frequently used — as a food diary.

A user could pull out their phone and snap a picture of the meal, then later add it to the food diary, where it would be automatically analyzed. The app would show the food items and estimated sizes, and the user could either accept the estimates if they seem correct or alter them if they're wrong. Over time the algorithms would get better at guessing items, sizes, and calories, too. Google has recently filed to patent the technology; no time frame for launch has been provided.

SOURCE: Popular Science