About EcoSort AI
Inspiration
Rapid urbanization has created massive waste management challenges in our cities. Poor segregation at source leads to contaminated recyclables, overflowing landfills, and increased pollution. We wanted to build a simple yet impactful AI solution that empowers citizens to sort waste correctly.
What We Built
EcoSort AI is an intelligent waste classification system that uses a camera and deep learning to identify waste types (plastic, glass, metal, paper, organic, trash) in real-time and provides instant disposal guidance.
The working demo is a web application built with Gradio where users can upload an image of waste and get immediate classification with confidence score and actionable advice.
How We Built It
- Model: Used a pre-trained ResNet50 model from PyTorch with transfer learning principles.
- Interface: Built a clean, user-friendly web app using Gradio.
- Development: Developed entirely in Google Colab for quick iteration and GPU support.
- Trained the model on the Garbage Classification dataset and created a shareable public demo.
Challenges Faced
- Initial model accuracy was low due to limited training time. We improved it by using transfer learning and freezing early layers.
- Balancing speed and accuracy for a one-day hackathon project.
- Making the UI intuitive and visually appealing for non-technical users.
What We Learned
- The power of pre-trained models for quick computer vision prototypes.
- Importance of good data preprocessing and UI/UX even in technical projects.
- How AI can directly contribute to environmental sustainability and smart city initiatives.
EcoSort AI demonstrates how accessible AI tools can be used to solve real urban problems. We're excited to take this from demo to a physical smart bin prototype in future rounds!
Log in or sign up for Devpost to join the conversation.