Much of this code was originally adapted from image compression algorithm by Balle et al as implemented by jonycgn (see: https://github.com/tensorflow/compression). New functions were added and existing ones modified as necessary within the 'balle' directory.
The most substantially modified elements, including much new structure for the network, are contained in the 'li combo'directory for reference.
The modified Mentzer code is also included for reference
Use:
Example: Compress an Image
'python bls2017.py compress example.png compressed.bin'
Example: Compress and Image for Li w/ Optional Extra Argument to Denote Image Map Instead of Image
'python li_001.py compress example.png compressed_map.bin map'
Example: Decompress an Image
'python bls2017.py decompress compressed.bin reconstructed.png'
Example: Train a model for baseline balle algorithm
'python bls2017.py train'
Example: Train a model for li implementation 001
'python li_001.py train'