Dual-path dehazing network with spatial-frequency feature fusion

Li Wang, Hang Dong, Ruyu Li, Chao Zhu, Huibin Tao, Yu Guo, Fei Wang

Abstract: Highlights•We propose a dual-path network to exploit the spatial-frequency knowledge.•A novel relaxation dense feature fusion (RDFF) makes the features more versatile.•A novel frequency attention (FA) enhances the representation of learned features.•On the popular datasets, our method makes new state-of-the-art performance.