A team of researchers from the University of Tel Aviv developed a neural network able to read a recipe and generate a image of what it would look like a finished product and cooked. As if DeepFakes was not serious enough, we can not be sure that the delicious food we see online is real.
The Tel-Aviv team, composed of researchers Ori El Bar, Ori Licht and Netanel Yosephian created their artificial intelligence with the help of a modified version of a contradictory network generative ( GAN ) called StackGAN V2 and picture 52K / recipe combinations of the gigantic recipe1M dataset .
The team has developed an artificial intelligence that can take almost any list of ingredients and instructions and determine the appearance of the finished food product.
[It] It all started when I asked my grandmother for a recipe for her legendary fish chops with tomato sauce. Due to her advanced age, she did not remember the exact recipe. So, I was wondering if I could build a system that would produce the recipe from an image of the food. After thinking about this task for a moment, I concluded that it was too difficult for a system to get an exact recipe with real quantities and with "hidden" ingredients such as salt, pepper, butter, flour, etc.
Then I wondered if I could do the opposite instead. Namely, generate food images from the recipes. We think this task is very difficult for humans to do, let alone for computers. Since most current AI systems are trying to replace human experts with easy tasks for humans, we thought it would be interesting to solve a type of task that even exceeds the capabilities of the human. As you can see, this can be done with some success.
Note that the image quality of the recipe1M dataset is poor compared to CUB and Oxford102 datasets. This results in many blurry images with poor lighting conditions, "porridge-like images" and the fact that the images are not square-shaped (which makes it difficult to model). This may explain the fact that both models have managed to generate images of "porridge-like" foods (eg, pasta, rice, soups, salads), but struggle to generate food-shaped images. distinctive (eg hamburger, chicken, beverages). ).
This is the only type of artificial intelligence we know, so do not expect it to be an application on your phone anytime soon. But the writing is on the wall. And, if it's a recipe, the artificial intelligence of the Tel Aviv team can turn it into an image that looks good enough that, according to the research paper, humans sometimes prefer it. a picture of reality.
What do you think?
The team intends to continue to develop the system, in the hope that it will expand to other areas than that of the company. ;food. Ori Bar El told us:
We plan to extend the work by training our system with the rest of the recipes (we have about 350,000 more images), but the problem is that the current dataset is of poor quality. We did not find any other datasets available that fit our needs, but we could build a dataset containing the text for children's books and the corresponding images.
These talented researchers may have cursed gourmets on Instagram to a world where we can not really know if what drools us is real, or if a robot gets a feel for the breath. ]