Unsupervised Monocular Depth and Ego-motion Learning with Structure and Semantics 之论文详解

1、论文详解

2、问题

1、如何理解inverse-warping

inverse-warp:具体如何实现，指的是什么？
从i到j的映射过程其实是，将j通过j的depth和ij之间的ego-motion映射到i上的位置，然后通过线性插值获取该j点的像素值，该像素值就是i映射到j的过程中所估计的值，与GT上的j值做loss来训练网络。

这是一个双线性插值的过程

2、cars moving in front at roughly the same speed often get projected into infinite depth

This is because the object in front shows no apparent motion, and if the network estimates it as being infinitely far away, the reprojection error is almost reduced to zero which is preferred to the correct case.

参考文献
（1）digging into self-supervised monocular depth estimation（CVPR 2018）（2）Every pixel counts: Unsupervised geometry learning with holistic 3d motion understanding(ECCV 2018)
本文提出的方法是如何解决的？