MIT has created a neural network for pizza analysis

A team of MIT researchers has recently developed an AI model capable of reading instructions and creating finished products. This AI model has very high applicability in areas like construction and robotics, but the team decided to experiment with something that we all love: making pizza .

PizzaGAN , the latest neural network created by geniuses at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) and Qatar Computer Research Institute (QCRI) , is an adversarial network. The ability to create images of pizza before and after being baked.

This neural network creates an image of a pizza before and after being baked.

If you're wondering why it's just an image, that's because it's just an initial experiment. When we hear about robots that can replace people in the food industry, we often imagine a machine from Boston Dynamics walking around in the kitchen, waving, cooking, baking a burger, fried potatoes, and exclaimed "Done" when finishing the cake, but the truth was much less exciting.

In fact, the restaurants described above use "automation" technology, not "artificial intelligence". Robots that make burgers don't care if it's a real cake or a hockey ball. It does not understand what a burger is, or whether the final product should look like. Those machines can be brought into the Amazon warehouse to take on the task of gluing tape items, no problem, because they are not smart.

On the contrary, what MIT and QCRI have done is to create a neural network capable of looking at the image of a pizza, determining the type of cake and the necessary ingredients, then studying how to place the ingredients. How to make each layer before putting the cake in the oven. It understands - in a way that any other AI understands what they are trained in - what the pizza making process is from start to finish.

The team has achieved this amazing result through a novel method of using modules. They develop AI with the ability to visualize pizza based on added or removed ingredients. You can give it a picture of a pizza being made, and ask it to remove mushrooms and onions, which will create a new cake image without those two.

According to the researchers: " From the visual aspect, each step of the instruction can be seen as a way to change the appearance of the dish by adding additional objects (eg adding an extra element ) or change the form of an existing dish (eg cooking a dish) ".

For robots or machines to make pizza in real life, it will have to understand what pizza is. And so far, the people we are - even the ultra-smart people at CSAIL and QCRI - have done a better job of re-creating the image of the food in the robot rather than tasting the food.

The AI ​​model behind PizzaGAN may be useful in many other areas.

For example, Domino Pizza is currently testing a computer vision solution to monitor every pizza coming out of the oven, to determine if they look good enough according to company standards. Things like cake topping density, maturity, and roundness can be measured and evaluated by machine learning in real time to ensure customers are not in the wrong hands.

The solution of MIT and QCRI integrates the pre-cooking step and determines the appropriate method of placing the cake to create a delicious pizza and flatter the eyes. At least that's in theory - we still have many years to have a complete solution to use AI from start to finish, from preparing, cooking, and serving pizza to guests.

Of course, pizza is not the only thing that a robot can do once it understands what ingredients, steps, and end-results should look like. Researchers conclude that the AI ​​model behind PizzaGAN may be useful in many other areas.

" Although we have only evaluated this model in pizza making, we believe a similar approach would be very promising in making layered other foods such as burgers, sandwiches, salads. In addition to food, our model can also be applied to other areas such as digital fashion shopping assistant, an area where the main activity is to combine different layers of clothing.