AdderIC: Towards Low Computation Cost Image Compression

Recently, learned image compression methods have shown their outstanding rate-distortion performance when compared to traditional frameworks. Although numerous progress has been made in learned image compression, the computation cost is still at a high level. To address this problem, we propose AdderIC, which utilizes adder neural networks (AdderNet) to construct an image compression framework. According to the characteristics of image compression, we introduce several strategies to improve the performance of AdderNet in this field. Specifically, Haar Wavelet Transform is adopted to make AdderIC learn high-frequency information efficiently. In addition, implicit deconvolution with the kernel size of 1 is applied after each adder layer to reduce spatial redundancies. Moreover, we develop a novel Adder-ID-PixelShuffle cascade upsampling structure to remove checkerboard artifacts. Experiments demonstrate that our AdderIC model can largely outperform conventional AdderNet when applied in image compression and achieve comparable rate-distortion performance to that of its CNN baseline with about 80% multiplication FLOPs and 30% energy consumption reduction.

1. INTRODUCTION

Image compression is one of the most essential tasks in both signal processing and computer vision fields. Recently, with the rapid development of deep learning, learned image compression methods [1], [2], [3], [4] have drawn much attention and shown their excellent performance when compared to traditional frameworks [5], [6], [7]. Although numerous progress has been made in learned image compression, the recently proposed models still maintain high computation complexity, being not conducive to deployment on devices with limited resources, e.g., mobile phones and embedded devices that have extensive needs of image compression.