Learning Monocular Depth in Dynamic Scenes via Instance-Aware Projection ConsistencyBest paper, Qualcomm Innovation Fellowship Korea 2020