Learning Accurate Entropy Model with Global Reference for Image Compression

TL;DR

This paper introduces a global reference-based entropy model for deep image compression that leverages both local and global context, significantly improving compression efficiency.

Contribution

It proposes a novel Global Reference Model that uses global context for better entropy estimation, along with a mean-shifting GDN module for enhanced performance.

Findings

Outperforms state-of-the-art methods in rate-distortion performance

Leverages global context for improved entropy modeling

Introduces mean-shifting GDN module for better results

Abstract

In recent deep image compression neural networks, the entropy model plays a critical role in estimating the prior distribution of deep image encodings. Existing methods combine hyperprior with local context in the entropy estimation function. This greatly limits their performance due to the absence of a global vision. In this work, we propose a novel Global Reference Model for image compression to effectively leverage both the local and the global context information, leading to an enhanced compression rate. The proposed method scans decoded latents and then finds the most relevant latent to assist the distribution estimating of the current latent. A by-product of this work is the innovation of a mean-shifting GDN module that further improves the performance. Experimental results demonstrate that the proposed model outperforms the rate-distortion performance of most of the state-of-the-art methods in the industry.