CIDBNet: A Consecutively-Interactive Dual-Branch Network for JPEG Compressed Image Super-Resolution

Xiaoran Qin, Yu Zhu, Chenghua Li, Peisong Wang, Jian Cheng

Abstract: Compressed image super-resolution (SR) task is useful in practical scenarios, such as mobile communication and the internet, where images are usually downsampled and compressed due to limited bandwidth and storage capacity. However, a combination of compression and downsampling degradations makes the SR problem more challenging. To restore high-quality and high-resolution images, local context and long-range dependency modeling are both crucial. In this paper, for JPEG compressed image SR, we propose a consecutively-interactive dual-branch network (CIDBNet) to take advantage of both convolution and transformer operations, which are good at extracting local features and global interactions, respectively. To better aggregate the two-branch information, we newly introduce an adaptive cross-branch fusion module (ACFM), which adopts a cross-attention scheme to enhance the two-branch features and then fuses them weighted by a content-adaptive map. Experiments show the effectiveness of CIDBNet, and in particular, CIDBNet achieves higher performance than a larger variant of HAT (HAT-L).