Guess_the_doodle

This project uses Convolutional Neural Networks (CNNs) for the task of Image Classification. The model looks at the picture of a doodle you drew and by extracting and processing all sorts of 'features', guesss the correct label with ~87% accuracy. The Quick Draw dataset from the Google Quick Draw game has been used for the purpose of training. This implementation classifies a doodle to one of the 60 output classes.

Demo

A video demonstration of this project can be found here

Running this code

You can run this application on your localhost. To do so, install python 3.6.x or 3.7.x, then setup the following directory structure.

project
  |- deps

cd to the project directory and clone. Then...

# Setup a python virtual environment
virtualenv -p /usr/local/bin/python3 deps/     #`/usr/local/bin/python3` is the path to your python installation
source deps/bin/activate

# Install the required dependencies.
# The `env_setup.sh` script takes care of this task.
chmod +x env_setup.sh
./env_setup.sh

# Start the Flask server.
python3 app.py

• After this, open your browser and visit http://localhost:5000 to enjoy playing. :)

Training your own model

• There is a Jupyter Notebook called doodle_classifier.ipynb in this repository. If you would like to train this model from scratch, simply open that notebook in Colab and start your work.

• Training would yield a .h5 file (called doodle_model.h5 by default). This file contains the information about the architecture and weights of the model. This is the file that is used by our demo web application to classify the user made doodles. To deploy your own model, just replace the doodle_model.h5 file in our demo directory with your own.