STanH : Parametric Quantization for Variable Rate Learned Image Compression

TL;DR

STanH introduces a differentiable, parametric quantizer for learned image compression, enabling flexible rate control without retraining multiple models, thus simplifying deployment and reducing storage needs.

Contribution

The paper presents STanH, a novel differentiable quantizer that allows a single model to adapt to multiple bitrates, improving efficiency over traditional fixed-rate models.

Findings

Achieves comparable compression efficiency to state-of-the-art methods.

Enables variable rate coding with a single trained model.

Reduces training and storage costs significantly.

Abstract

In end-to-end learned image compression, encoder and decoder are jointly trained to minimize a $R + {\lambda}D$ cost function, where ${\lambda}$ controls the trade-off between rate of the quantized latent representation and image quality. Unfortunately, a distinct encoder-decoder pair with millions of parameters must be trained for each ${\lambda}$, hence the need to switch encoders and to store multiple encoders and decoders on the user device for every target rate. This paper proposes to exploit a differentiable quantizer designed around a parametric sum of hyperbolic tangents, called STanH , that relaxes the step-wise quantization function. STanH is implemented as a differentiable activation layer with learnable quantization parameters that can be plugged into a pre-trained fixed rate model and refined to achieve different target bitrates. Experimental results show that our method enables variable rate coding with comparable efficiency to the state-of-the-art, yet with significant savings in terms of ease of deployment, training time, and storage costs