This repository is an exercise project using a sequence variational auto-encoder (VAE) to encode and generate MNIST hand-written digits.
A simple script to interpolate between latent representations of 2 mnist digits is also included.
Other than python built-in modules, the following are additional required dependicies
- numpy
- keras==2.2.2
- tensorflow==1.10.0
- matplotlib==2.1.2
The model is implemented in file model.py, and the code to train is written in train.py.
python train.py # pass -h option for more optionsAfter running train.py, trained model is saved in saved_models/
To interpolate between digits, run
python interpolate.py # pass -h option for more options