Compression without quantization is a lossy transform coding framework for compression an arbitrary real-valued tensor x.
This repo presents an application of this framework to lossy image compression. A fully convolutional Probabilistic Ladder Network is implemented in Tensorflow (TF), Tensorflow Probability (TFP) and Tensorflow Compression (TFC) that can be trained and used on an arbitrary image dataset.
The code is writtent in Python 3.6.
Clone this repository
> git clone https://github.com/gergely-flamich/compression-without-quantization.git
> cd compression-without-quantization
Create a virtual environment and activate it
> virtualenv cwoq-venv
> source cwoq-venv/bin/activate
Install the requirements (note that this may take a while)
(cwoq-venv)> pip install -r requirements.txt
To train / compress / decompress and to see what options are available for these, simply run
(cwoq-venv)> python code/miracle.py --help
Please contact the author at flamich.gergely@gmail.com