eatsafe

Is it a potato?
logo

Inspiration

Have you ever had a new food offered to you, ask what it is, and just have your friend say "I'm not sure what's in here, but it's good!" For some of us, we can ignore our worries and consume the new food, but for millions of people worldwide, serious allergies impede their freedom to eat. Trying new foods without an ingredient list or previously obtained knowledge becomes a serious health hazard.

This didn't feel fair to us. Meet eatsafe.

What it does

We solved this by building a complete platform for users to photograph their food, then immediately check whether it contains common ingredients. By allowing computers to disambiguate dishes, users can have more confidence than ever when eating.

How we built it

We built this tool by utilizing deep learning. We trained a neural network on thousands of hand-labelled pictures. Rather than simply scraping tens of thousands of data-points, we focused on obtaining a smaller sample size, but with a higher quality. We utilized an Inception-V3 model architecture, with a dynamic softmax output layer.
This approach allowed us to easily retrain our model for both our 'potato / not-potato' task and for out 'potato / meat / neither' task!

Challenges we ran into

We originally wrote our code to run entirely through TensorFlowJS, but after developing the code for hours we found that serving the model with a Python solution would be more complete.