How many countless hours have you spent scrolling through Pinterest trying to figure out a recipe to make with your limited ingredients? Drooling over food bloggers' mouthwatering photography, the food search often still requires a trip to the store. When money's tight and you've had a long day, a trip to the grocery store is out of the question. If only there was a way to turn photos into recipes. Wait, hold on, now there is.
MIT researchers trained an artificial intelligence system to suggest recipes based on a photo of ingredients. The system called Pic2Recipe identifies the ingredients in the photo and uses a database called Recipe1M that contains over 1 million recipes to create a modern dinner aide. The research team lead by MIT graduate Nick Hynes developed this system in hopes of gaining a better understanding on dietary habits.
Guess what's on the menu today??? I call it my internal hugging ? ??definitely a grey Monday essential! ???? #love #instalove #instafood #hanoicaphe #ingredients #internalhugging #vietnamese #streetfood #photography #pho #authentic #delicious #healthy #noodles #broth #fresh #food #cleaneating #vegan
The Computer Science and Artificial Intelligence Laboratory at MIT developed the Recipe1M database by taking previous research and combining it with information from popular food recipe websites such as AllRecipes. By doing so, they are attempting to create the most accurate large-scale datasets in food recognition using corresponding ingredients.
Previous datasets have been achieved in predicting food accuracy from image recognition, but only to a certain point. According to MIT News, Swiss researchers developed "an algorithm that could recognize images of food with 50 percent accuracy." The problem in achieving a success rate of 100 percent accuracy may lie in the size of the dataset.
The CSAIL researchers and officials have utilized research of these types of datasets to build their own, and thus training a neural network to recognize patterns in food images. By identifying the patterns, the network can then make connections with valuable insight that results in correct recipe suggestions, even those with ambiguous foods.
While the system did very well on baked good recipes, it had a difficult time recognizing others like sushi rolls. The team plans on working out the kinks, with a goal of better understanding food preferences and dietary preferences. Over time, researchers may even turn Pic2Recipe into a meal aide that will make cooking easier for the home chef.
It's important to keep in mind that deep-learning and machine learning will result in different variations over time as the system develops to take in every food item possible.
As Hynes told MIT News,
"This could potentially help people figure out what's in their food when they don't have explicit nutritional information. For example, if you know what ingredients went into a dish but not the amount, you can take a photo, enter the ingredients, and run the model to find a similar recipe with known quantities, and then use that information to approximate your own meal."
HELLO summer breakfast forever. ✨ . . . http://www.howsweeteats.com/2016/06/summer-breakfast-tacos/ . . . #f52grams #fwx #buzzfeast #eeeeeats #feedfeed #foodblogfeed #foodandwine #beautifulcuisines #bhgfood #marthafood #RSlove #thekitchn #bareaders #foodblogeats #howisummer #ABMfoodie #howsweeteats #mywilliamssonoma #hautecuisines #chowhound
A project such as this could be a game changer in home cooking. Social media is flooded with images of food that cause stomachs to grumble and mouths to water. The problem is, there's no recipe. By taking our obsession with food photography and turning it into an edible meal, Pic2Recipe could change how we view food posts on social media.
Food photography into real food recipes? Sounds like there's going to be an app for that. Until then, you can test out the demo page here.