Decipher Bengali Graphemes

This project employs Convolutional Neural Networks to decipher Bengali graphemes, as part of the Bengali.AI challenge on Kaggle, https://www.kaggle.com/c/bengaliai-cv19.

Report

A report of our most important findings is available at Report Bengali Competition.pdf.

Retrieve data

The data can be downloaded directly from Kaggle to any/most systems. This required the package kaggle to be installed, which can be done with pip install kaggle. On Google Colab this is installed by default.

To retrieve the data, the username and API key need to be provided. These can be obtained from Kaggle.com -> account -> Create New API Token, which downloads a json file with the username and key. Set these before running the script with os.environ['KAGGLE_USERNAME'] = 'username from file' and os.environ['KAGGLE_KEY'] = 'kaggle key from file'.

Finally, retrieve the data with the command python CNN/fetch_data.py, which can be executed from a Jupyter Notebook using !python CNN/fetch_data.py or from Python code using os.system('python CNN/fetch_data.py').

This gives:

os.environ['KAGGLE_USERNAME'] = 'username from file'
os.environ['KAGGLE_KEY'] = 'kaggle key from file'
os.system('python CNN/fetch_data.py')