STanH : Parametric Quantization for Variable Rate Learned Image Compression

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Presta, Alberto, Tartaglione, Enzo, Fiandrotti, Attilio, Grangetto, Marco
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910646016147456
author Presta, Alberto
Tartaglione, Enzo
Fiandrotti, Attilio
Grangetto, Marco
author_facet Presta, Alberto
Tartaglione, Enzo
Fiandrotti, Attilio
Grangetto, Marco
contents In end-to-end learned image compression, encoder and decoder are jointly trained to minimize a $R + λD$ cost function, where $λ$ 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 $λ$, 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
format Preprint
id arxiv_https___arxiv_org_abs_2410_00557
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle STanH : Parametric Quantization for Variable Rate Learned Image Compression
Presta, Alberto
Tartaglione, Enzo
Fiandrotti, Attilio
Grangetto, Marco
Computer Vision and Pattern Recognition
Multimedia
In end-to-end learned image compression, encoder and decoder are jointly trained to minimize a $R + λD$ cost function, where $λ$ 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 $λ$, 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
title STanH : Parametric Quantization for Variable Rate Learned Image Compression
topic Computer Vision and Pattern Recognition
Multimedia
url https://arxiv.org/abs/2410.00557