In this paper, we propose to enhance learned image compression systems with a richer probability model for the latent variables. Previous works model the latents with a Gaussian or a Laplace distribution. Inspired by binary arithmetic coding , we propose to signal the latents with three binary values and one integer, with different probability models. A relaxation method is designed to perform gradient-based training. The richer probability model results in a better entropy coding leading to lower rate. Experiments under the Challenge on Learned Image Compression (CLIC) test conditions demonstrate that this method achieves 18% rate saving compared to Gaussian or Laplace models.
@article{arxiv.2002.09259,
title = {Binary Probability Model for Learning Based Image Compression},
author = {Théo Ladune and Pierrick Philippe and Wassim Hamidouche and Lu Zhang and Olivier Deforges},
journal= {arXiv preprint arXiv:2002.09259},
year = {2020}
}