This repository implements a computer-vision framework for real-time waste recognition and classification. The system is built in PyTorch using a ResNet-18 convolutional neural network to identify and categorize common trash types, supporting automated or assisted recycling workflows.
The model was trained on a combination of open-source waste classification datasets sourced from Kaggle and evaluated on diverse physical trash samples, achieving approximately 95% classification accuracy across varied lighting and object orientations. The pipeline integrates image preprocessing, supervised training, and inference optimized for practical deployment.
This project was developed as part of a recycling-focused startup prototype at Wharton’s Global Youth Innovation & Startup Camp, where the model was evaluated not only for technical performance but also for feasibility in real-world adoption. The implementation prioritizes deployability and contamination reduction in recycling streams rather than theoretical optimization, emphasizing applied computer vision in environmental systems.