Learning robust and scale-aware monocular depth estimation (MDE) requires expensive data annotation efforts. Self-supervised approaches use unlabelled videos but, due to ambiguous photometric ...