VGGNet for modified MNIST dataset

Repository for deep learning image recognition (on a modified MNIST dataset)

Final Setup

The below setup instructions are outdated, and only run a custom CNN which achieves low (27%) accuracy on the training set for unknown reasons. You only need the FINAL_VGG.ipynb file - we run it in Google Colab with their GPUs. This final model achieved a ~95% accuracy on the modified MNIST dataset during training, and 96% on the held-out Kaggle test set.

Setup

1. Clone this repository
2. Start a virtual environment

python3 -m venv env
source env/bin/activate

3. Install project dependencies

pip install -r requirements.txt

4. Run the driver file