The aim of this project is to build on current works in lossy Compressive Autoencoders (CAE).
The file 'Simple CAE MNIST.py' translates the work found from [1] into PyTorch and modifies the model, mainly a convolutional layer, to improve performance.
The orginal code from [1] using a stored format of 16 neurons yielded the following results:
The modified model had the following results for 16 neurons:
[1] https://blog.paperspace.com/autoencoder-image-compression-keras/