Can’t decide which meal to order?
Why not let your little kids draw something they would like to have for this dinner?
With the power of TensorFlow.js, we created a machine learning (ML) model that can recognize more than 40 dishes. The challenge of this project was to gather as varied drawing styles as possible in a short period of time. Undoubtedly, the input quantity was essential, so was the data diversification. To achieve both aspects in one go, we developed a mobile data collecting application that made collecting and labelling data effortless. Just use our web app to capture a drawing then choose one of the predefined classes to label it.
In close collaboration with Grab and Google, over 10,000 doodles were submitted from across the countries. Our ML model used the collected data to learn what each dish should look like. In addition, we also generated drawing data using image processing techniques and the Generative Adversarial Networks (GANs) model to create more variety of drawing styles, which helped enhance the accuracy of our ML model. Finally, we combined the model with our web app, and voila, Delivery Doodles is now ready to deliver this enjoyable experience to more and more families!
Delivery Doodles will magically turn your kids’ doodles into food orders!