Sebica: Lightweight Spatial and Efficient Bidirectional Channel Attention Super Resolution Network

TL;DR

Sebica is a lightweight network that incorporates spatial and efficient bidirectional channel attention mechanisms that significantly reduces computational costs while maintaining high reconstruction quality and demonstrates significant improvements in real-world applications, specifically in object detection tasks, where it enhances detection accuracy in traffic video scenarios.

Abstract

Single Image Super-Resolution (SISR) is a vital technique for improving the visual quality of low-resolution images. While recent deep learning models have made significant advancements in SISR, they often encounter computational challenges that hinder their deployment in resource-limited or time-sensitive environments. To overcome these issues, we present Sebica, a lightweight network that incorporates spatial and efficient bidirectional channel attention mechanisms. Sebica significantly reduces computational costs while maintaining high reconstruction quality, achieving PSNR/SSIM scores of 28.29/0.7976 and 30.18/0.8330 on the Div2K and Flickr2K datasets, respectively. These results surpass most baseline lightweight models and are comparable to the highest-performing model, but with only 17% and 15% of the parameters and GFLOPs. Additionally, our small version of Sebica has only 7.9K parameters and 0.41 GFLOPS, representing just 3% of the parameters and GFLOPs of the highest-performing model, while still achieving PSNR and SSIM metrics of 28.12/0.7931 and 0.3009/0.8317, on the Flickr2K dataset respectively. In addition, Sebica demonstrates significant improvements in real-world applications, specifically in object detection tasks, where it enhances detection accuracy in traffic video scenarios.

Figures (5)

• Figure 1: Comparison of the SR images derived from listed networks.
• Figure 2: The attention module comprises 6 attention blocks, each integrating a spatial and an efficient bidirectional channel attention mechanism. Additionally, we propose a smaller version called Sebica_small, which consists of 4 attention blocks. Spatial attention and channel attention mechanism are discussed in Section \ref{['sec:spatial_attn']} and Section \ref{['sec:channel_attn']} separately.
• Figure 3: Spatial attention block. To minimize computation while maintaining performance, we use dimension-wise average and max operations instead of traditional average and max pooling.
• Figure 4: Efficient bidirectional channel attention block. Each cubic in various color represents one channel. Standard Sebica configures total 16 channels, we setup 8 channels for Sebica_small version.
• Figure 5: Comparison of the SR images derived from baseline.