This paper proposes one Fully Convolutional Network of Self-supervised Monocular Depth Estimation with Global Receptive Field and Unreasonable Matching Penalty