Inspiration

We were inspired by the HooHacks prompt on data science regarding training an image classifier to differentiate between recyclable and compostable items.

What it does

Our web application shows alarming facts, visualization, links to a slides presentation, and runs inferences through the pre-trained A.I. model MobileNet.

How we built it

We built using the JavaScript frameworks of React and ONNX. We also used Google Colaboratory notebooks with Python, Tensorflow, and fast.ai.

Challenges we ran into

The biggest challenge was figuring out how to deploy our A.I. model to a React web application. In the end, unfortunately, we were not able to deploy the model we specifically trained, but we were able to get a pre-trained model (MobileNet V2) up and running on our live web application.

Accomplishments that we're proud of

  • Training a ~95% accurate trash classifier to label images as either ['cardboard', 'glass', 'metal', 'paper', 'plastic', 'trash']
  • Actually being able to deploy a pre-trained neural network onto a web application

What we learned

  • How to deploy an A.I. model to a web application
  • How to use fast.ai to train an image classification model
  • General machine learning concepts (e.g. difference between validation data set and testing data set)

What's next for HooHacks-2021-Recycle

  • Practically, to figure out how to get our actual trained model onto the React application.
  • Our long-term vision is to allocate robots to search through the USA, sort through trash bins, and place any misclassified trash into their proper bins (e.g. if a compostable item is inside a trash bin, then put that item in a compostable item). In addition, these robots could pick up any litter and put litter into their proper bins as well.

GitHub link

https://github.com/jacobsomer/HooHacks-2021-Recycle

Sources

Prizes Elgibility

  • Best Beginner Hack (for three of us here, it's our first hackathon)
  • Best Data Science Hack

Built With

Share this project:

Updates