Artificial intelligence taught to restore recipes from pictures of finished dishes

Allegedly, the development has surpassed the conventional search engine.
The artificial intelligence experts at Facebook have created a system that can restore a recipe by analyzing a photo of it. The analysis takes only a few seconds, and the result is a list of ingredients and a list of steps necessary to create a dish.

Although the system cannot find out from the picture of a specific pie what kind of flour is used in it and what the exact temperature of the oven was when baking, the recipe prepared by the system will give a very similar (and tasty) approximation to the original.

The system relies on computer vision to extract information from a digital image. It includes two neural networks: one defines the ingredients, and the other develops the recipe, more precisely, finds a suitable one in the database, which uses a subset of the Recipe1M database (the developers left about 350,000 of the 1 million recipes).

Note that in this way, AI predicts the presence of even those ingredients in the recipe that are not visible in the photo, say, sugar or cinnamon. For this, knowledge of which ingredients often appear together is used.

Allegedly, the system surpasses the regular search engine both in identifying ingredients and in creating plausible recipes. Although the development was of a research nature, it is part of a wider project aimed at teaching computers to see and understand the world.