Some question about the result

[Chenlei6]

In the training of the KITTI dataset, disparity estimation results will not be constrained or punished in some areas without Ground-Truth. When you are training, does the predicted result have outliers in the Ground-Truth-free area? If so, how did you solve it? I hope to hear from you.

[WASCHMASCHINE]

I cannot speak for the author, but I can confirm that there are strong artifacts for some parts of the scene that are not covered by the Lidar ground truth. For example, look at the Autobahn / highway scenes and notice the disparity estimates in the sky.

[Chenlei6]

Thanks for your reply！