Exploring Structural Sparsity in Neural Image Compression

TL;DR

This paper introduces an adaptive binary channel masking method to induce sparsity in neural image compression networks, enabling significant computational reduction and acceleration with minimal quality loss.

Contribution

It proposes a simple plug-in method for channel importance assessment and pruning, improving efficiency without hardware changes.

Findings

Up to 7x reduction in computation

Achieved 3x acceleration in inference

Negligible performance drop during compression

Abstract

Neural image compression have reached or out-performed traditional methods (such as JPEG, BPG, WebP). However,their sophisticated network structures with cascaded convolution layers bring heavy computational burden for practical deployment. In this paper, we explore the structural sparsity in neural image compression network to obtain real-time acceleration without any specialized hardware design or algorithm. We propose a simple plug-in adaptive binary channel masking(ABCM) to judge the importance of each convolution channel and introduce sparsity during training. During inference, the unimportant channels are pruned to obtain slimmer network and less computation. We implement our method into three neural image compression networks with different entropy models to verify its effectiveness and generalization, the experiment results show that up to 7x computation reduction and 3x acceleration can be achieved with negligible performance drop.