Keras autoencoders (convolutional/fcc) [proof of concept]

This is an implementation of weight-tieing layers that can be used to consturct convolutional autoencoder and simple fully connected autoencoder. It might feel be a bit hacky towards, however it does the job.

Convolutional autoencoder example

Run conv_autoencoder.py.

FCC autoencoder example

Run conv_autoencoder.py. Conv layer (32 kern of 3x3) -> MaxPool (2x2) -> Dense (10) -> UpSample (2x2) -> DeConv layer (32 kern of 3x3)

FCC autoencoder example

Run fcc_autoencoder.py. Conv layer (32 kern of 3x3) -> MaxPool (2x2) -> Dense (10) -> UpSample (2x2) -> DeConv layer (32 kern of 3x3)