This Program Can Tell You the Ingredients Behind Every Meal By Its Picture

Researchers at the Massachusetts Institute of Technology (MIT) have good news for all food lovers. Their recent paper gives us information about a program they have developed which can tell you the ingredients of a dish just by analyzing its image.

How it Works

The program called Recipe1M, was fed about 800,000 images of food along with a million recipes from a handful of cooking websites. The program analysed the images similar to how Google Photos App works. It ‘learns’ which image belongs to which recipe and can then recognize similar images.

Although the software is not foolproof, it can still guess ingredients correctly 65% of the time though the accuracy of the outcome greatly depends on the type of food.

It can identify dessert ingredients like eggs, butter and flour easily. But when given tougher ‘less defined’ foods like milk shakes or sushi, the program somewhat struggles. Though, it should be noted that even for us humans with our great eyesight and inference, it is difficult to judge ingredients of such blended cuisines.

Its Implications

This remarkable use of AI can open up some interesting doors to the future. In today’s age, people are becoming more and more diet conscious and closely observe their calorie intake. Recipe1M can be used to do just that. You could easily keep track of the ingredients you’re ingesting and cut down on those which make you fat.

If you’re more on the skinny side, then you can look for foods which have high contents of proteins and fats.

Another slick use would be to look for the ingredients of a course you’ve just been served for the first time in your life. You can maybe try the recipe when at home.

Earlier Attempts

This isn’t the first time someone has tried such a feat. In 2014, a group of Swiss researchers developed their own variation of a cooking program dubbed ‘Food-101’. It could also identify food but was much less accurate.

This seems to be what’s holding AI development back in food based applications. There just aren’t large enough datasets present to make the required predictions. But Recipe1M seems to have overcome that obstacle with its huge library of images and recipes. Giving it a much better chance at guessing the ingredients, and consequently the recipes, correctly.

The AI based software will only get better from this point on as the developers add to its volume of data.

What do you think about this piece rather unique use of AI? Would you be interested in using it? Let’s discuss it in the comments below.