Simplifying waste disposal with AI-powered image classification.
https://athulsubashm.github.io/Ecolens
⚠️ Notice: Hugging Face models used by this demo are no longer available, so the app does not function.
Navigating waste disposal, especially with varied recycling streams, proved to be a common challenge (we even struggled for 5 minutes to sort a single item on campus!). This highlighted a significant opportunity for improvement. Further research revealed that current recycling processes often rely on manual sorting and expensive equipment. We saw a clear opportunity for AI-driven image classification and robotics to enhance efficiency in recycling centers, and this vision inspired the creation of Eco Lens.
In its current mobile web application form, Eco Lens allows users to classify objects by uploading an image or taking a picture with their camera. Our AI image detection model identifies the object, and this information is then processed by an AI text classifier to determine its disposal category (e.g., "plastic bag" falls into "non-recyclable"). Crucially, the app also provides clear instructions on how an average person can recycle or dispose of the item themselves.
The application was developed using the React.js framework. For image identification, we integrated the powerful google/vit-base-patch16-224 Vision Transformer model from Hugging Face. Text classification and generation of disposal information are powered by the llama3-8b-8192 model from Groq. Our collaborative development workflow was managed on GitHub, and the web application is hosted on GitHub Pages, linked to a domain from GoDaddy.
As this was the first hackathon for all three team members, and our first time utilizing AI models, we faced several learning curves. A significant challenge was attempting to train our own AI model from scratch; consequently, we opted to leverage powerful pre-trained AI models instead.
We successfully integrated AI models into our application for the first time. We also managed to establish a collaborative development environment and create a robust deployment workflow, all accomplished within a compressed 2-day timeframe.
We gained valuable experience in utilizing AI models and integrating APIs into web applications. We also honed our teamwork skills, developing a fully functional prototype application within 48 hours, complete with a polished, elegant, and user-friendly UI.
We envision two main paths for Eco Lens's future development:
- Train our model to significantly increase accuracy.
- Integrate data feeds to a robotic arm to enable real-time waste sorting.
- Seek access to a recycling center to gather real-time data for generating a robust training dataset.
- Implement location-based services to provide relevant disposal center information:
- Example: For a laptop, direct users to electronic disposal centers.
- Example: For clothing, suggest local thrift stores or textile recycling centers.