Whether the sky area is involved in the loss calculation?

[WeijianZhang123]

Hello authors,
You said in the paper that the GT of the sky area is set to 0. Does this part of the GT participate in the loss calculation?
I know that invalid regions are also set to 0 in some public datasets.
Looking forward to your reply！

[LiheYoung]

Yes, the sky (inverse depth = 0) is included in the loss computation.

[WeijianZhang123]

Yes, the sky (inverse depth = 0) is included in the loss computation.

Thanks for your quick reply! I'd like to ask one more question.
If the detected sky area already has ground-truth, is it also set to 0? Because I noticed that there were minor errors in sky detection, such as detecting a white foreground object as a sky.

[LiheYoung]

We set the detected sky region as 0, only if it originally does not have any label.