As high as an app that can tell you a piece of pizza is not actually a hot dog (thanks Jian Yang) set the bar for becoming a Shazam for food, the folks at MIT set their sights a little higher.
It’s far from perfect, but researchers at MIT have designed a system that draws on a database of more than 101,000 images of food to help identify a dish and then match it with a recipe.
Created using a food-identifying algorithm called the Food-101 Data Set, the Pic2Recipe system still has issues with some images, but is able to correctly identify and finding a matching recipe about 65% of the time, Techcrunch reports. (We tried it — nailed pizza; missed waffles).
“It’s an issue of getting the scale correct,” researcher Nick Hynes told the publication. “When people take pictures of food, there’s a lot of variation in style: whether taken from close up or far away; whether it’s of a single item, multiples, part of a complete dish. Of course, that’s not unreasonable, since even a human might think a single cookie is a pancake when zoomed out.”
In general, the system has a good deal of success identifying pastries and other baked goods but struggles with complex dishes like sushi rolls and foods with indecipherable ingredients.
If you’d like to check it out for yourself, here’s the demo page. Bon appétit.