Pyramid Convolutional Network for Single Image Deraining

Jing Zhao, Jiyu Xie, Ruiqin Xiong, Siwei Ma, Tiejun Huang, Wen Gao; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 9-16

Restoring images corrupted by rain streaks is important for many computer vision applications in outdoor scenes. Benefiting from the fast inference and excellent feature representation capability, deep convolutional neural networks (CNN) have achieved significant performance improvement for image deraining and attracted considerable attention recently. However, for the images with complex background, the performance of these CNN-based methods is still unsatisfactory. Addressing this issue, we develop a new pyramid convolutional neural network, which is composed of multiple subnets, for image deraining, and name it PDRNet. To take full advantage of multi-scale redundancy, the network decomposes the rainy images into multi-scale subbands via a hierarchical wavelet transform and then process them by several sub-networks respectively. In particular, wavelet transform also plays the role of downsampling and enlarges the receptive field without increasing depth or sacrificing efficiency of network. Experimental results show that our PDRNet can not only achieve promising deraining performance quantitatively and qualitatively, but also benefit high-level computer vision tasks.