Handwriting MNIST dataset

This repository contains code for training a model to generate this dataset. If you are interested in how this was created please look at my post where I described the process.

How to preview?

To preview simply run this command:

python show.py

How to train model?

1. Go to page: https://edwin-de-jong.github.io/blog/mnist-sequence-data

2. Download digit-images-thinned.tar.gz to data directory

3. Unpack it in data directory (you should end up with: data/digit-images-thinned)

4. Run this command from data directory:

python prepare_data.py

This will create pickle file with thinned MNIST dataset.

5. Now you can run training from base directory:

python train.py

You can run tensorboard for some visualizations.

6. After training if over you need to convert thinned dataset to handwritten dataset:

python convert.py --model=path_to_model --save=mnist-handwritten
                          ^ex. summary/experiment-0/models/handwriting-model-0

7. Preview dataset:

python show.py --dataset=mnist-handwritten.pkl