FOODI

Inspiration

My inspiration for this project was teaching people how to cook food and knowing the type of food they eating

What it does

This lens using SnapML multi classification model to distinguish between food types then shows the recipe for the detected food. Not only that, the lens can teach you how to cook more than 100 dishes

How we built it

I first started making this lens by training the M L model provided by Snap A-R and have used a food dataset found on kaggle. The results were great and I was able to get an accuracy rate of more than 95 percent. Of course, there were some challenges in the work, such as splitting the data and the long time it takes for the model to complete the training process. After that, I collected more than 100 recipes for well-known dishes, and that was the difficult task, as I decided not to collect the data manually, so I created a code using Node js to take the data and arrange it in a certain way so that I can use it within Lens Studio easily. After that, I imported the M L model into Lens Studio, where there is an amazing Snap M L tool that enables you to link the model in the work environment easily; What an amazing software! Don't you agree with me? The next step was to write a code that displays recipes based on detected food by the M L model; I faced some challenges here but I overcame them. I also used a 3D object controller, which is a helper provided by the Snap A-R that enables you to move 3d objects by touching them through the screen. I also used chain physics to add some fun to the 3D models. I used Head binding M L to stick some 3d objects on the head and face.

Challenges we ran into

Collecting 101 food recipes. Mapping the recipe to the detected dish.

Accomplishments that we're proud of

I am happy that this is the first time someone make this type of food lenses, and I am so proud that I can teach someone how to cook through the camera.