SC-Depth provides three versions (V1, V2, V3) of self-supervised learning approaches for monocular depth estimation from video sequences. Version 1 introduced geometry consistency loss and self-discovered masks for scale-consistent predictions. Version 2 added an auto-rectify network (ARN) to handle large rotational motions in hand-held camera footage. Version 3 improved robustness for highly dynamic scenes and sharper depth at object boundaries using external pretrained depth networks.