The computer learned to determine the calorie content of dishes from the photo

German researchers have created an algorithm that can determine the calorie content of dishes, as well as their content

At the same time, the algorithm does not refer to existing databases of the nutritional value of products in real time.

“N + 1” writes briefly about the development.The developers tried to solve a problem that may concern people losing weight: if you prepared a dish yourself or ordered it in a cafe, it is quite difficult to determine the calorie content and nutritional value. Existing algorithms for this task simply recognize food in photos and then find matches in databases. This approach is not very accurate.

The authors of the new algorithm used a different method.It consists in allowing the neural network to calculate the nutritional value based on the image in one step. The neural network was trained on a large ImageNet dataset, which was formed using recipe sites and a database of nutritional values ​​of various products.

As a result, the algorithm can determine not only the calorie content of the dish and the content of proteins, carbohydrates fats, but also guess the ingredients.