Block-optimized Variable Bit Rate Neural Image Compression

TL;DR

This paper presents a block-based neural image compression system with novel techniques for binarization, variable bit rates, entropy optimization, and normalization, demonstrating incremental performance improvements.

Contribution

It introduces multiple innovations including binarization simulation, multi-network variable bit rate control, entropy-friendly representations, and inference-stage code optimization.

Findings

Incremental performance gains from each proposed contribution

Effective variable bit rate control with multiple networks

Enhanced compression efficiency through entropy-friendly representations

Abstract

In this work, we propose an end-to-end block-based auto-encoder system for image compression. We introduce novel contributions to neural-network based image compression, mainly in achieving binarization simulation, variable bit rates with multiple networks, entropy-friendly representations, inference-stage code optimization and performance-improving normalization layers in the auto-encoder. We evaluate and show the incremental performance increase of each of our contributions.