• 

  Screenshot of the Working Camera React App

Compostable

Inspiration

• Composting is a sustainable practice

What it does

• A mobile app that takes pictures and classifies objects as noncompostable vs compostable
• We setup our trained classification model as an API so the mobile app can easily send a POST request with the camera image and receive a json response with the predicted object class

How we built it

• React Native + Expo for the mobile app
• PyTorch for Object classification ML
• Figma for wireframing
• Exploring PyTorch Live to deploy ML models to a mobile app
• Used dataset from: https://www.kaggle.com/sapal6/waste-classification-data-v2

Challenges we ran into

• Integrating the mobile app with the classification model was difficult as we could not get the API requests to send the image properly from the react app

Accomplishments that we're proud of

• Learned how to use react to create one of our first apps
• Trained a resnet18 PyTorch model on ~ 20k images using Amazon Sagemaker Studio Lab cloud vms with a GPU
• Achieved ~ 80% accuracy on the test dataset after just 10 epochs (this took 30 mins!)

Built With

• pytorch
• react-native