How qubitpro/recyclewaste works
Project type and purpose
• This is an ML-powered image classification system for household waste, built with transfer learning (TensorFlow/Keras backbones) and exposed via a Gradio web application deployed on Hugging Face Spaces.
• Core use case: train and benchmark CNNs on a local dataset, then serve the best model for real-time inference through a browser interface.
File structure
• Dataset/ (six class folders: cardboard, glass, metal, paper, plastic, trash)
• Week 1/ and Week 2/ (exploratory/training notebooks)
• Final Notebook.ipynb (consolidated data prep, training, evaluation)
• app.py (Gradio inference script)
• requirements.txt (dependencies)
• README.md (project description, live demo link, results summary)
README for additional insights
• Multiple CNN backbones benchmarked (EfficientNet variants, MobileNetV2, ResNet50)
• Transfer learning with ImageNet weights
• Class-weighting for imbalanced data
• Visualization of metrics (training curves, confusion matrix)
• Deployed serverless on Hugging Face Spaces via Gradio