We propose a semi-supervised learning framework for monocular depth\nestimation. Compared to existing semi-supervised learning methods, which\ninherit limitations of both sparse supervised and unsupervised loss functions,\nwe achieve the complementary advantages of both loss functions, by building two\nseparate network branches for each loss and distilling each other through the\nmutual distillation loss function. We also present to apply different data\naugmentation to each branch, which improves the robustness. We conduct\nexperiments to demonstrate the effectiveness of our framework over the latest\nmethods and provide extensive ablation studies.\n