This is a small, self-contained repo that lets you compress your own images using some of the latest and greatest (pretrained) neural compression algorithms. It uses the excellent compressai library.

Set-up

Clone the repo and create a conda environment by running

conda env create -f env.yml
conda activate compress

To run the example do

python3 compress_my_image.py --image=test_image.dng --quality=5 --plot

The model used to compress an input image is that of Ballé et al (2018) as implemented and trained in the compressai library. The utility of this repo is in the data preparation (which is not hard but time-consuming -- it took me half a day to set up). This includes things like:

• loading an RGB image (8-bit integers) and converting it in the appropriate format (tensor of floats in [0, 1])
• ensure appropriate padding for the scale hyperprior architecture to work out
• calculating basic statistics (size, bits per pixel, distortion)
• converting the reconstruction back to RGB space
• conda environment setup etc.

Disclaimer: use at your own risk

References

Ballé, J., Minnen, D., Singh, S., Hwang, S. J., & Johnston, N. (2018). Variational image compression with a scale hyperprior. ICLR. arXiv:1802.01436.

Bégaint, J., Racapé, F., Feltman, S., & Pushparaja, A. (2020). Compressai: a pytorch library and evaluation platform for end-to-end compression research. arXiv preprint arXiv:2011.03029.