This app aims to identify recipes from photographs

Seeing a photo of a gorgeous dish is nice, but it won’t do much good in learning the recipe – at least for now. Researchers have developed an algorithm that locates a recipe for a dish using only a photograph. 

The algorithm has been made into an app called Pic2Recipe (if they had asked me, I would have suggested Pic2Plate). To get it to work, researchers provided the computer with a database of a million recipes, along with a photograph of the final dish. The recipes and images came from allrecipes.com and food.com. You can see the database at the Pic2Recipe website.

To use the app, you upload a picture of the food for which you want the recipe. The deep learning algorithm analyzes the picture and attempts to link it to the correct recipe.  Usually you will get several recipes; the number depends on the confidence level of the system that it has found the correct food. 

The system is far from perfect – currently it only has about a 65% success rate. Variables that affect the accuracy include whether the photo is a close-up or not, or if there are other food items in the photograph besides the one you are searching for.

It’s an interesting concept but I’m not sure how valuable it would be – it is not very often that I see a photo of a dish and want the recipe without knowing the name of the dish. If I know what it is called I will just search EYB to find the recipe. It might be good when you are at a restaurant, smell a dish at a nearby table, and want to know what it is. In that case, however, getting a photo of the dish might prove to be tricky unless the diners are very friendly. What do you think of this concept?