Character recognition CNN with server based interface.
The CNN model was trained on EMNIST dataset. My best result on the dataset after trying several different setups was 88% accuracy. Results on my own handwriting were around 60%.
Python 3.5.4, packages from requirements.txt
and graphviz.
To create/load CNN model and train it run: python nerual_network.py
.
NOTE: When creating the model you might experience lags, these should only last for few minutes until the training starts.
To start the interface run: python manage.py runserver 0.0.0.0:8000
.
You can then access the interface by visiting http://localhost:8000.
If you want to try different settings for the CNN or change filepaths, change the values in config.py
.