🎨 AI Doodle Classifier

A real-time web app that classifies hand-drawn doodles using TensorFlow.js.

🚀 Try Live Demo

Features

• Real-time AI predictions as you draw
• 365 object categories
• Mobile-friendly with touch support
• No backend required

Run Locally

git clone https://github.com/deshan0/doodle-classifier.git
cd doodle-classifier
python -m http.server 8000

Open http://localhost:8000

Note: Requires local server due to CORS

🛠️ Tech Stack

• Frontend: HTML5 Canvas, CSS3, Vanilla JavaScript
• AI/ML: TensorFlow.js CNN model (15MB)
• Dataset: MNIST, MNIST-FASHION, & Google's Quick Draw (365 classes)
• Model: Convolutional Neural Network trained on 28x28 grayscale images

📊 Technical Details

• Input: 280x280 canvas → preprocessed to 28x28 grayscale
• Model: CNN with ~1M parameters
• Classes: 365 categories from 3 different datasets
• Browser Support: Chrome, Firefox, Safari, Edge (ES6+ required)

Drawing Tips

• Draw simple, clear shapes
• Use full canvas space
• One object per drawing

Built with ❤️ by Deshan