Use sustAInably to help you recycle your plastic waste with AI!

🔗 try sustAInably here

• Capture or upload a plastic code image
  
• Install progressive web app (PWA) for quick access
  
• Look up disposal guidelines for items
  
• Use AI to recycle smarter
  
• Track your recycled plastics
  

The model was trained on images representing the seven plastic resin codes. The dataset uses images from this Kaggle dataset, which is also available in ml/seven_plastics.

The model is built using TensorFlow's EfficientNet for transfer learning, accelerating the learning process. Training was conducted on a GPU-powered machine using Python. The training script is available in ml/train.py, where various architectures and parameters were tested before finalizing the model.

To integrate real-time predictions with the front end, the model was converted for compatibility with TensorFlow.js. Using Web Workers, predictions run smoothly without affecting the app’s performance.

When an image is uploaded, the app processes it into a tensor, and the model predicts the most probable resin code, displaying relevant recycling information.

To improve accuracy, users can provide feedback on incorrect predictions. This benefits both the user and the model:

1. Users receive correct recycling information.
2. Model performance in real-world scenarios is monitored.
3. New data (if permitted) helps refine future versions.

Although the feedback feature is present in the front end, it isn't connected to a backend to keep the app lightweight and privacy-friendly.

• Kaggle Dataset – Plastic resin code images
• Collletttivo – Mattone font
• Stubborn – Illustrations
• Unsplash – Additional images

• TensorFlow – Model training and inference
• React Camera Pro – Camera integration