contador Skip to content

Google is now able to count calories from food photos

Im2Calories able to calculate the calories of a dish by analyzing the photo. on food the most advanced artificial intelligence project by Google

Food and calories are at the center of the most advanced Google search. Artificial Intelligence (AI) at the base of the Google search engine is, increasingly, at the center of the company's strategy, which for this reason in 2014 acquired DeepMind – a $ 14 billion for the development of Deep Learning. But Google does not stop there, it has in fact integrated robotics, autonomous driving cars and delivery drones into its organic organization.

Food and calories in the study of artificial intelligence

But what are Google's AI experts working on lately? In a word: FOOD.At the Rework Deep Learning Summit held in Boston, Google researcher Kevin Murphy unveiled a project related to the highly sophisticated deep learning algorithms that analyze static images of food and estimate the calories present in that dish or food. Im2Calories the system that, by looking at an image, counts the foods present and can give an idea of ??the calories based on the size of the individual foods and the condiments or sauces associated with them. But c more, Im2Calories does not require high resolution images, the Instagram format, for example, perfectly suitable for analysis.

Count the calories in a simple way

The purpose of Im2Calories is to simplify the preparation of a food diary, identifying the foods without being manually entered into an app. a semi-automated system, which can be officially launched when food identification accuracy of 30% is reached, because thanks to large-scale use the system will be able to optimize itself quickly. The problem of obesity and junk food ( junk food sold at ridiculous prices, remains a scourge in the US and a commercial version of Im2Calories could favor the popularity of the system and an improvement in the eating habits of many and like many other deep learning applications, it combines visual analysis with recognition schemes.

Im2Calories learns to recognize food and becomes more precise in counting calories

Im2Calories able to associate a range of calories related to the appearance of a food, and was created with a profound learning ability, which can therefore improve through frequent use. The purpose of the many deep learning systems is to minimize the quantity of time needed to provide data to improve performance and user experience.

Estimate the calories through the pixels

Im2Calories focuses on the pixels of an image, which is why, the greater the number of images it will have, the more accurate will not only be the recognition of food, but also the caloric estimates. Despite the lack of extreme accuracy at the moment, according to Murphy , Im2Calories will know how to prove revolutionary. For me it is obvious that people will consider this tool useful. Probably the calorie estimate will be reduced by at least 20%, but it doesn't matter. It will be a matter of months, weeks or years, but we will be able to increase the accuracy significantly. And now we can start thinking about merging information from a large number of people so we can compile a more precise statistic. I have colleagues working in the health sector and everyone is interested in this solution.

Im2Calories, a patented system

Google has recently submitted Im2Calories a patent license and Murphy has preferred not to share details on the official release. If we are able to apply this technology to food, then Google will be able to apply it to an infinite series of areas, such as the location of cars, the counting of vehicles on the road, the diversification of the cars identified and to determine which direction they are traveling. We could start doing traffic analysis or forecast parking. And since everything starts from the data, the technology will be unchanged, only the data will change.Food Calories from IM2Calories photos