Memory Oriented Transfer Learning for Semi-Supervised Image Deraining

Figure 1: Image rain removal in the real world. PReNet [23] and Syn2Real [34] are among the state-of-the-art supervised and semi-supervised methods, respectively. All the methods are trained on DDN-SIRR [29]. The top and bottom rows show that MOSS can remove more rain streaks while preserving better background textures (Note that the missing traffic markings may cause serious accidents in autonomous driving).