Auto-encodeur optimisé au sens débit-distorsion : indépendant de la quantification?

Thierry Dumas 1 Aline Roumy 1 Christine Guillemot 1
1 Sirocco - Analysis representation, compression and communication of visual data
Inria Rennes – Bretagne Atlantique , IRISA-D5 - SIGNAUX ET IMAGES NUMÉRIQUES, ROBOTIQUE
Abstract : This work relates to image compression via a transform learned by an auto-encoder. It tries to adapt the quantization to this transform instead of fixing it. We propose to jointly learn the transform and the quantization. Moreover, we analyze whether different quantization steps can be applied to a transform learned for one step only. We show that the second approach corrects the aw of the state-of-the-art auto-encoder for image compression: having to learn one transform per compression rate.
Complete list of metadatas

Cited literature [7 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-01579257
Contributor : Thierry Dumas <>
Submitted on : Wednesday, August 30, 2017 - 5:28:01 PM
Last modification on : Friday, September 13, 2019 - 9:50:02 AM

File

quantification_ae.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01579257, version 1

Citation

Thierry Dumas, Aline Roumy, Christine Guillemot. Auto-encodeur optimisé au sens débit-distorsion : indépendant de la quantification?. GRETSI 2017, Sep 2017, Juan-les-Pins, France. ⟨hal-01579257⟩

Share

Metrics

Record views

570

Files downloads

171