For any of the tech giants that dominate the business landscape today, investments in R + D + i represent an important starting each year. It is therefore not surprising that Facebook, the largest social network in the world, have a team specially dedicated to the development of tools based on artificial intelligence.
This is the FAIR, acronym Facebook’s Artificial Intelligence Research, and among its recent achievements is a breakthrough that could prove a double-edged sword for restaurateurs.
We talk about a software developed by Adrianna Romero and laboratory equipment Montreal (Canada) whose purpose is generate recipes from the image of the dish.
It looks like science fiction, but it's not. If we put a picture of a steak and chips garnished interpret the pixels to figure out what meat is, how it was prepared and the approximate amounts shown on the plate. Then a written recipe all sorts of details about the best way to prepare the menu is created.
Who says a meat dish, says a soup, a salad or any other dish we can imagine. The true potential of the image recognition system lies in its versatility when it comes to identify with certainty what was reflected in the photograph, and its ability to issue a text good yarn, organized, cohesive, consistent and relevant which can be used to reproduce the dish.
The tool has been designed to meet the needs of users in social networks. How many times have we seen an appetizing meal and ask how to do it has not given us answer? Certainly, in the case of foodies curious, They will be more than you can count on hands.
The head of the research department laboratory Montreal, Joelle Pineau, expressed view about this tool follows: "Everyone takes pictures of your meals today, sometimes there are ingredients that you can see, but there are also ingredients that are not always, as salt, the sugar, and other things like that".
Those ingredients that are not apparent in the picture itself are those who hide one of the best kept secrets of image recognition tool. Not that the software I guess what leads and what not dish. The truth is that the system has been trained with a huge number of cases resolved so far has become able to associate the image with an existing prescription or a combination thereof. This method is what is commonly known as machine learning, one of the pillars of the AI and neural networks.
A) Yes, one of the fears that had been identified among the restorers is belied. Had heard voices crying out that artificial intelligence could reveal the secret of the dishes design, but if those dishes are not listed anywhere else and are in fact the product of imagination and knowledge of a chef inspired, not fear. Would dawn if the cook in question would have us think that their preparations are more special than they really are. In any case, disregarding rare cases, These concerns are totally unfounded, restorers to base their business model innovation and irreproducibility their menus can be calm.
further, Pineau was quick to indicate that the development of this tool is merely a theoretical exercise. For long interest AI system arising between users of Facebook (and also between Instagram, they have wasted no time in demanding similar applications for its platform to make progress), no social network intends to implement this method of image recognition At this precise moment.
The real reason that has created the tool is because "We need to have machines that understand the world. Understand not only what is visible in the world, but understand that if you have a cake is very likely to have sugar in it ', in the words of Pineau own.
In the same line of work they are immersed in the Institute of Technology Masachussetts (OF. UU.).
In July 2017, several researchers Computer Science and Artificial Intelligence Laboratory (CSAIL), in collaboration with colleagues at the Institute for Computational Research Qatar (QCRI), They published a study entitled Crosswise integration with learning recipes and food images in which they would display their software based on AI, Pic2Recipe, powered by a database called Recipe1M containing one million prescriptions and 800 000 pictures of food.
The results of the neural network were still poor, reported as the BBC, identified spaghetti bolognese as a bowl of noodles with "fat" and Italian tomato sauce. Where we have taken leave of translation 'sebum', read unidentifiable lump of fatty substance. It is less palatable but more faithful to the recipe proposed by the demo technique. Clearly much remains to be done.
In any case we suspect a future in which this tool were to be used for the name of the dishes shown in a photo, or to give an approximate but realistic recipe same. The possible application to break down language barriers is equally interesting. A app identifying the mobile dish served us well for a trip abroad and that gives us information about it? Yes, please!
Some of the restaurant owners who use professional photographs of their menus to attract clientele Facebook o Instagram they could even automatically generate these recipes (and subsequently retouch if necessary) thus increasing the interest and involvement of users passing through the business profile.
We will have to wait while the team of postdoctoral researchers Antonio Torralba (WITH) and the FAIR Laboratorio Joelle Pineau (Facebook) develop more precise tools. It is quite possible that the near future there are real world applications for restoration.