Noticed people struggling to find something to throw away.
Wondered if AI could help people sort waste correctly in real-time.
Decided to build an ML/AI app that detects trash and classifies it into categories.
Gathered datasets like TrashNet to get images of paper, plastic, metal, glass, and compost.
Collected more real-world images to improve the dataset and balance the classes.
Used data augmentation to simulate lighting changes, blur, and object angles.
Trained a convolutional neural network using TensorFlow and Keras.
Fine-tuned MobileNetV2 for mobile-friendly performance and decent accuracy.
Achieved over 85% accuracy in classifying common waste items.
Built a cross-platform mobile app using Flutter.
Integrated the trained model using TensorFlow Lite for real-time performance.
Allowed users to take or upload a picture to get instant classification.
Displayed the correct bin: landfill, recycle, or compost.
Faced challenges with blurry or dirty items that confused the model.
Found that ambiguous items like greasy cardboard or plastic-covered paper were difficult.
Learned how important clean, well-labeled data is for machine learning.
Struggled to compress the model for phones without losing too much accuracy.
Learned to balance model performance and app responsiveness.
Realized that UI/UX is just as important as backend logic for helping users.
Designed the app to be fast, simple, and educational.
Want to expand the app to include barcode scanning and voice prompts.
Hope to deploy it in schools, malls, and public spaces.
Believe this project can help reduce contamination in recycling streams.
Excited to keep improving the app with better models and more user feedback.
Log in or sign up for Devpost to join the conversation.