Abstract: Transparent glass or mirrors, which are very common in daily life, often bring severe challenges to computer vision tasks. Therefore, this paper presents an efficient glass segmentation model, in which a multi-scale contextual convolution (MSCC) module is proposed to mining the glasses features in various scales and remain the feature map's origin resolution. Experimental results demonstrate that the proposed model achieves significant performance boost compared with state-of-the-arts, illustrating the outstanding contextual features representing ability.