Binary Probability Model for Learning Based Image Compression

Abstract

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 gradientbased 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.

Sparse Flow Adversarial Model For Robust Image Compression

“…By defining the number of nodes at the end of an encoding subnet, the image is transformed into a feature vector with a small number of elements, thus reducing the redundancy of the image. So far, various kinds of deep learning networks, including recurrent neural networks (RNNs) [5,6], convolutional neural networks (CNNs) [7,8,9,10,11,12,13], and generative adversarial networks (GANs) [14,15], have been explored for image compression. Although these methods have achieved great results in certain datasets, there are still some shortcomings.…”