Spatially-Adaptive Learning-Based Image Compression with Hierarchical Multi-Scale Latent Spaces

TL;DR

This paper introduces a hierarchical multi-scale latent space in neural image compression, improving rate efficiency and complexity management over single-scale models, with potential for faster decoding.

Contribution

It extends state-of-the-art neural compression with a second hierarchical latent space and a gain unit, achieving 7% rate savings and marginal complexity increase.

Findings

Outperforms traditional autoencoders by 7% rate savings.

Adds minimal complexity despite extra latent space.

Potential for decreased decoding time.

Abstract

Adaptive block partitioning is responsible for large gains in current image and video compression systems. This method is able to compress large stationary image areas with only a few symbols, while maintaining a high level of quality in more detailed areas. Current state-of-the-art neural-network-based image compression systems however use only one scale to transmit the latent space. In previous publications, we proposed RDONet, a scheme to transmit the latent space in multiple spatial resolutions. Following this principle, we extend a state-of-the-art compression network by a second hierarchical latent-space level to enable multi-scale processing. We extend the existing rate variability capabilities of RDONet by a gain unit. With that we are able to outperform an equivalent traditional autoencoder by 7% rate savings. Furthermore, we show that even though we add an additional latent space, the complexity only increases marginally and the decoding time can potentially even be decreased.