CNSNet: A Cleanness-Navigated-Shadow Network for Shadow Removal

The key to shadow removal is recovering the contents of regions with guidance non-shadow regions. Due inadequate long-range modeling, CNN-based approaches cannot thoroughly investigate information from To solve this problem, we propose a novel cleanness-navigated-shadow network (CNSNet), shadow-oriented adaptive normalization (SOAN) module and shadow-aware aggregation transformer (SAAT) based on mask. Under mask, SOAN formulates statistics region adaptively applies them for region-wise restoration. SAAT utilizes mask precisely guide restoration each shadowed pixel by considering highly relevant pixels shadow-free global pixel-wise Extensive experiments three benchmark datasets (ISTD, ISTD+, SRD) show that our method achieves superior de-shadowing performance.