Reviewers like the many positive aspects of this paper, including good results, solid experiments, and sound geometric principles. All reviewers recommend accept after rebuttal. AC agrees with this consensus recommendation. In the camera ready version, the authors should revise the paper as discussed in the rebuttal. In particular, the authors should improve the clarity of the presentation. In addition, the authors should discuss the remaining issue raised by R2 regarding why the network can gradually correct the mistakes it make in the occlusion and depth estimation.