This app has all the ingredients to be successful.
The team at MIT CSAIL has created Pic2Recipe, a new AI-driven software that can spit out ingredients and cooking directions for a dish just by analyzing a picture of it.
Think Shazam for food -- kind of like that app the guys from HBO's "Silicon Valley" tried building.
How does Pic2Recipe work?
CSAIL researchers borrowed from cooking websites to build a database of over 1 million recipes and then connected it to an aggregate of modified versions of pre-existing food recognition platforms.
The team then trained a neural network to link a picture that’s uploaded to it with one in its database of food images and, eventually, a recipe.
During tests, the software was able to bring up the correct recipe 65 percent of the time, its creators said, but even then it sometimes left out ingredients.
Pic2Recipe right now does best with desserts like cookies or muffins, MIT explained in a release about it, and so far has most difficulty determining exact ingredients for more ambiguous foods like sushi rolls and smoothies.
Researchers said they will continue improving the accuracy and abilities of the software, possibly even bringing a “dinner aide” feature to it that will examine the ingredients in your fridge and tell you the dishes you can make with them.
You can try Pic2Recipe right now at its MIT CSAIL demo page.
Though its creators have no plans to commercialize Pic2Recipe, its techniques are expected to be made available for free, so others with more of an entrepreneurial calling can bring it to app stores. We think we might know a team who would have been interested: