An artificial intelligence (AI) project called Im2Calories has been unveiled in the Rework Deep Learning Summit at Boston last week. Google is developing an app based on this project that is capable of analyzing a still picture of food, and measure how many calories each food item has, in relation of the food size to the plate, as well as any condiments. The program does not require high resolution captured images, and any standard Instagram-quality shot would be accepted.
Kevin Murphy, Google research scientist, stated the AI technology analyzes the depth of pixels in a picture, and utilize "sophisticated deep-learning algorithms" to judge the shape and size of a food item. Im2Calories uses visual analysis partnered with pattern recognition, and its algorithms would evolve and get better as more people use the app to provide more accurate calorie information. Im2Calories can draw connections between what a given piece of food looks like, and vast amounts of available caloric data. With this program, the users would not be burdened by encoding each food item to the app as well as guess other variables like the serving size of the food.
Google has recently issued an official request for the patent of Im2Calories. Murphy has kept most of the important details and its availability, but Murphy is more excited about the long-term goals that this AI technology could achieve. He mentioned that if this would be a successful app for food, the program could be applicable for other situations such as traffic pattern analysis, wherein through deep-learning data accumulated over time, then it could be used to predict vacant parking spots.
Obesity has been a major crisis in the United States and a commercial version of Im2Calories would be hugely popular for the diet-conscious individuals, as well as for food tracking and improved recommendations by nutritionists and dieticians. It would be a very good business move for Google to invest in this technology, most especially that it could also be utilize for their autonomous robot cars, which could futuristically predict empty parking spot within minutes.
A Google representative Jason Freidenfelds mentioned algorithms for Im2Calories were still at their research stage, and that there were no definite product plans as of the moment.
"We don't have more to say at this point on whether or how this might make its way into some future product(s), because we'd really just be speculating," Freidenfelds stated.
Photo: Alpha | Flickr
