Tested on Python 3.6.7
Model accuracy when trained on EMNIST balanced dataset: 78.14% NOTE: The default training parameters were used. To get better results, you can try changing the training parameters or maybe some pre-processing.
- Create a virtual environment
- Install dependencies:
pip install -r requirements.txt
The model uses the extended MNIST dataset. To setup the dataset, run python dataset.py. To know the CLI options, append it with -h or --help.
To train the model, run python model.py. To know the CLI options, append it with -h or --help. The model is saved at models/classifier.joblib.
Run python drawer.py. It will open a window. Draw the characters over there and press ESC key to close the window. It will then run predictions on the characters drawn and publish result to console and display the image in a window with it's title being the predicted label. Press ESC key to close the window and run predictions on next image.
