Shadow Removal using GTA Road Dataset

In this paper, we propose a end-to-end Road Shadow Removal Network (RSRNet) on GTA road scene. Proposed network consists of shadow detection part and removal part. The shadow detection network separately predicts edges and regions of the shadow to accurately predict shadow masks. Given the mask, the shadow removal network removes shadows by predicting parameters of the shadow region between shadow free and shadow. The RSR network effectively removes the shadow while preserving the non-shadow region information. We evaluate proposed network quantitatively and qualitatively to confirm the performance on shadow removal in complex scenes.