This implementation slightly adjusts the convolutional autoencoder implementation found here: https://github.com/mikesj-public/dcgan-autoencoder
The main adjustments are:
- Altering the net input size
- Adding additional convolutional layers to the front of the autoencoder
- Toggling Pooling before MSE
- Adjusting the encoding cost multiplier
- Adding functionality for printing the entire test set
- Adjusting dataprocessing.py to create a masked image dataset
The instructions for running the code are copied below from the original implementation:
I assume knowledge of IPython (Jupyter), pip and virtualenv (not complicated to learn if not). The following should work on unix systems. Working in a virtualenv, run
pip install -r /path/to/requirements.txt
You should download the CelebA dataset from website (you're looking for a file called img_align_celeba.zip). Unzip into this directory then run
./dataprocessing.py
This will crop the images to the right size and store them in HDF5 format.
Next run the dcgan notbook.