QVRF: A Quantization-error-aware Variable Rate Framework for Learned Image Compression

TL;DR

QVRF introduces a novel framework for learned image compression that enables a single model to efficiently operate over a wide range of variable bitrates by controlling quantization error, outperforming existing methods.

Contribution

The paper proposes QVRF, a quantization-error-aware framework that achieves wide-range variable rates in learned image compression without significant performance loss or additional parameters.

Findings

QVRF enables continuous variable rates within a single model.

QVRF outperforms current variable-rate methods in rate-distortion performance.

QVRF maintains performance with minimal additional parameters.

Abstract

Learned image compression has exhibited promising compression performance, but variable bitrates over a wide range remain a challenge. State-of-the-art variable rate methods compromise the loss of model performance and require numerous additional parameters. In this paper, we present a Quantization-error-aware Variable Rate Framework (QVRF) that utilizes a univariate quantization regulator a to achieve wide-range variable rates within a single model. Specifically, QVRF defines a quantization regulator vector coupled with predefined Lagrange multipliers to control quantization error of all latent representation for discrete variable rates. Additionally, the reparameterization method makes QVRF compatible with a round quantizer. Exhaustive experiments demonstrate that existing fixed-rate VAE-based methods equipped with QVRF can achieve wide-range continuous variable rates within a single model without significant performance degradation. Furthermore, QVRF outperforms contemporary variable-rate methods in rate-distortion performance with minimal additional parameters.