why are visible regions blacked out in visibility maps?

[xiyichen]

Thank you for the code release and the detailed documentations! I ran the visibility_batch script, but found the results a bit confusing. In this particular view for example (subject 0002, view 050), both feets are visible, but are marked invisible in the visibility map. I wonder why is it like this? Thanks for your response in advance!

[xiyichen]

I'm also unable to reproduce the example you showed in dataset.md by rendering the same view of the same subject.

In your visibility map, the left foot is partially observed, which makes sense:

But in mine, it's not observed at all, which does not align with the viewpoint:

[xiyichen]

I changed the smplx scale in visibility_batch.py from 180 to 100 in this line and was able to reproduce your visibility map.

I think this might be the issue since the scan_scale is already applied to the camera parameters, so having 180 here again would make the mesh too big and thus gets the feet cropped out of the image.

[zjh21]

I encountered a similar problem and my solution is the same as yours, setting scale to 100. I don't really know why the scale was set to 180 in the original code.

I changed the smplx scale in visibility_batch.py from 180 to 100 in this line and was able to reproduce your visibility map. I think this might be the issue since the scan_scale is already applied to the camera parameters, so having 180 here again would make the mesh too big and thus gets the feet cropped out of the image.