A simple web-based doodle predictor trained on top of the Google Quick, Draw! dataset. Draw something in your browser, and the model will try to guess what it is.
| Frontend | Backend | Model |
|---|---|---|
| Built with HTML5, CSS3, and vanilla JavaScript | Powered by FastAPI, providing a fast, async REST API | Developed using TensorFlow — a convolutional neural network trained on the Quick, Draw! dataset |
| Features a responsive drawing canvas with smooth real-time input handling | Accepts and processes base64-encoded canvas images for prediction | Designed to classify simplified grayscale doodles and output the most probable label |
Lightweight and efficient prediction endpoint at /predict |
This project uses a small Convolutional Neural Network (CNN) to predict doodles drawn by the user.
-
Trained on 50,000 doodles from the Quick, Draw! dataset
-
Total parameters: ~1.36 million
The model isn't super accurate yet since it was trained on a small part of the full dataset (50,000 doodles, accounting for only ~4%). But it still works well for common, simple drawings.