How does the "Hardware Occlusion Queries" culling work?

[John_Alcatraz]

@Kalle-H;673430

Occlusion queries render bounding boxes of objects after normal scene is rendered. Then query simply ask would bounding box mesh rendering resulted any visible pixels. So test is done against precise depth buffer. Bounding box of object is conservative so it won’t ever result false positives.(expect when camera moves and renderer use previous frame data)

Thanks!

It would be nice if it would work like that, but for some reason it doesn’t as far as I can tell.

This is a different test case:

https://puu.sh/uokDO/e05566f0a5.png

There is the ground, which is just a stretched cube. Then there is the cone, which is just a cone. And then there is the white wall, which is also just a stretched cube. You see that the wall cube is quite big in Z.

Now take a look at this gif:

https://puu.sh/uokzF/1cf28e5191.gif

First I am looking at the wall. Then I’m turning the camera so that it looks straight down, only the ground is visible. I’m calling “FreezeRendering” and looking back up at the wall. And you see that the wall is not occluded by the ground, even though it was completely hidden behind the ground when I called “FreezeRendering”. The cone is not rendered, so the cone was occluded correctly. If the wall would count as occluded once no pixel of its bounding box is visible, then it would not have been rendered. But it is rendered, and I don’t understand why