Deep Multiple Description Coding by Learning Scalar Quantization

TL;DR

This paper introduces a deep learning framework for multiple description coding that adaptively learns scalar quantizers and encoders to improve image compression efficiency and diversity, outperforming existing methods.

Contribution

The proposed framework uniquely integrates auto-encoder networks, entropy estimation, and end-to-end scalar quantizer learning for enhanced multiple description coding.

Findings

Outperforms several state-of-the-art methods in coding efficiency.

Effectively learns scalar quantizers and importance maps.

Generates diversified descriptions with structural dissimilarity loss.

Abstract

In this paper, we propose a deep multiple description coding framework, whose quantizers are adaptively learned via the minimization of multiple description compressive loss. Firstly, our framework is built upon auto-encoder networks, which have multiple description multi-scale dilated encoder network and multiple description decoder networks. Secondly, two entropy estimation networks are learned to estimate the informative amounts of the quantized tensors, which can further supervise the learning of multiple description encoder network to represent the input image delicately. Thirdly, a pair of scalar quantizers accompanied by two importance-indicator maps is automatically learned in an end-to-end self-supervised way. Finally, multiple description structural dissimilarity distance loss is imposed on multiple description decoded images in pixel domain for diversified multiple description generations rather than on feature tensors in feature domain, in addition to multiple description reconstruction loss. Through testing on two commonly used datasets, it is verified that our method is beyond several state-of-the-art multiple description coding approaches in terms of coding efficiency.