Cascade Wide Activation Multi-Scale Networks for Single Image Super-Resolution

摘要

In recent years, convolutional neural network (CNN) has made great progress in the field of single image super-resolution (SR). There is a strong correlation between multiple scales in SR, which has not been fully utilized by the previous works. In this paper, we first propose a new network structure based on WDSR, which is called JWDSR. JWDSR takes full advantage of the features generated by the residual blocks and leads to a better Peak Signal to Noise Ratio (PSNR). We then propose a cascaded multi-scale JWDSR network called CMDSR. CMDSR fully employed the correlation between multi-scale networks and supports multi-scale SR calculations. Experiment results on benchmark data sets show that CMDSR can get higher PSNR faster with fewer parameters than the previous works.