ScribbleNet

A tool which classifies hand-drawn doodles into distinct classes using a CNN (Convolutional Neural Network).


Image Image Image

ScribbleNet

Published on the Play Store

  • ScribbleNet is a Python package which is bundled as a REST API using Flask and deployed on Heroku.
  • It uses a Convolutional Neural Network (CNN) to recognize and classify hand-drawn doodles (source).
  • Trained on the QuickDraw dataset by Google (achieved a validation accuracy of 96%).
  • Built and Published an Android Application (Google Play) which communicates with the same API (in progress).
  • The web front-end is built using vanilla JavaScript, HTML and CSS.

Tech Stack

  • Flask (Python) - Rest API
  • Tensorflow
  • Keras
  • OpenCV
  • Android (Java)

GitHub Repository


Demonstration

Draw a Doodle in the canvas and ScribbleNet will attempt to guess what you drew.

Browser not supported.
  • Top 3 Predictions

  • *Prediction 1*
  • *Prediction 2*
  • *Prediction 3*