Conventional self-supervised monocular depth prediction methods are based on a static
environment assumption, which leads to accuracy degradation in dynamic scenes due to the
mismatch and occlusion problems introduced by object motions. Existing dynamic-object-
focused methods only partially solved the mismatch problem at the training loss level. In this
paper, we accordingly propose a novel multi-frame monocular depth prediction method to
solve these problems at both the prediction and supervision loss levels. Our method, called …

Method Training WxH The lower the better The higher the better Abs Rel Sq Rel RMSE
RMSE log δ< 1.25 δ< 1.252 δ< 1.253 Zhan FullNYU [43] Sup 608 x 160 0.130 1.520 5.184
0.205 0.859 0.955 0.981 Kuznietsov et al.[16] Sup 621 x 187 0.089 0.478 3.610 0.138 0.906
0.980 0.995 DORN [6] Sup 513 x 385 0.072 0.307 2.727 0.120 0.932 0.984 0.995
Monodepth [7] S 512 x 256 0.109 0.811 4.568 0.166 0.877 0.967 0.988 3net [29](VGG) S
512 x 256 0.119 0.920 4.824 0.182 0.856 0.957 0.985 3net [29](ResNet 50) S 512 x 256 …