EcoSort AI uses cutting-edge transfer learning to instantly identify waste types from your camera — and tells you exactly how to dispose of them correctly.
From a photo to a complete disposal guide in under 3 seconds. No account needed. No internet required after first load.
Use your device camera to take a live photo or upload an existing image. The scanner accepts any waste item — bottles, cans, food scraps, paper and more.
Your image is processed by MobileNetV2 — a deep learning model trained on 1452 images across 5 categories — running entirely in your browser using TensorFlow.js.
Instantly see the waste category, confidence score, step-by-step disposal tips, environmental impact, and estimated CO₂ saved by recycling correctly.
Earn EcoPoints for every correct scan. Level up from Eco Beginner to EcoSort Legend. Track your full recycling history and environmental impact over time.
Our model identifies five major waste categories found in everyday Pakistani households and workplaces — the types that matter most for local recycling.
Bottles, containers, bags, packaging material
Check & RecycleNewspapers, cardboard boxes, cartons
RecyclableCans, tins, aluminium, scrap steel
Highly RecyclableBottles, jars, glass containers
RecyclableFood scraps, fruit peels, garden waste
CompostableEcoSort AI isn't a rule-based system. It uses genuine deep learning — the same technology behind Google Photos — trained specifically for waste classification using Transfer Learning.
Built on MobileNetV2 pre-trained on 1.4 million images by Google. Only 164,613 parameters were trained — just 6.8% of the total model weight.
MobileNetV2Converted to TensorFlow.js using the SavedModel pipeline. The AI runs entirely in your browser using WebGL acceleration — no server required at all.
TensorFlow.jsYour photos never leave your device. All AI processing happens locally. Scan history and EcoScore data is stored only in your browser's localStorage.
On-Device AIEcoSort AI is a Progressive Web App. Install it directly from your browser — it works offline, launches instantly, and feels like a native mobile app.
PWA ReadyTrained on 1164 images, validated on 288 unseen images. EarlyStopping automatically halted training at the best epoch to prevent overfitting.
Paper achieves the highest F1 score (0.92). Organic shows high recall (0.96) but lower precision — expected because it was the smallest class with only 129 training images. Plastic and Metal sometimes confuse each other due to similar shiny visual surfaces.
Every correct scan earns EcoPoints. Points accumulate across sessions and unlock higher levels — making recycling a habit, not a chore.
Every correct scan builds better recycling habits. Join the green revolution — one item at a time.