SUBSTITUTIONAL NEURAL IMAGE COMPRESSION

被引：1

作者：

Wang, Xiao ^{[1
,2
]}

Ding, Ding ^{[2
]}

Jiang, Wei ^{[2
]}

Wang, Wei ^{[2
]}

Xu, Xiaozhong ^{[2
]}

Liu, Shan ^{[2
]}

Kulis, Brian ^{[1
]}

Chin, Peter ^{[1
]}

机构：

[1] Boston Univ, Boston, MA 02215 USA

[2] Tencent Amer, Palo Alto, CA USA

来源：

2022 PICTURE CODING SYMPOSIUM (PCS) | 2022年

关键词：

Image Compression; Online Training; substitution;

D O I：

10.1109/PCS56426.2022.10018005

中图分类号：

TM [电工技术]; TN [电子技术、通信技术];

学科分类号：

0808 ; 0809 ;

摘要：

We describe Substitutional Neural Image Compression (SNIC), a general approach for enhancing any neural image compression model, that requires no data or additional tuning of the trained model. It boosts compression performance toward a flexible distortion metric and enables bit-rate control using a single model instance. The key idea is to replace the image to be compressed with a substitutional one that outperforms the original one in a desired way. Finding such a substitute is inherently difficult for conventional codecs, yet surprisingly favorable for neural compression models thanks to their fully differentiable structures. With gradients of a particular loss back-propogated to the input, a desired substitute can be efficiently crafted iteratively. We demonstrate the effectiveness of SNIC, when combined with various neural compression models and target metrics, in improving compression quality and performing bit-rate control measured by rate-distortion curves.

引用

页码：97 / 101

页数：5