Dynamic Occlusion System

[DIFTOW]

Here are the culling methods listed in order they are applied to each primitive, which is also from least to most expensive in terms of rendering thread cost.

• Distance culling
• View frustum culling
• Precomputed occlusion culling
• Dynamic occlusion culling (hardware occlusion queries)

...
Dynamic Occlusion System

The dynamic occlusion system tracks visibility states for each primitive component and issues hardware occlusion queries based on heuristics to reduce the number of queries but increase culling effectiveness. The queries are read back from the GPU one frame later, so there is a one frame delay before an object is determined visible when coming around a corner, which can cause some popping. Occlusion queries are issued against the scene's depth buffer so both dynamic and static primitives can occlude, along with both opaque and masked materials. The dynamic occlusion system does not require any user setup to work, it is fully automatic.

I'm quite disappointed in UDN being so vague about many of it's features. The only thing said about the system is what it is, and not a word about how to enable/disable it.
If this feature indeed exists, it must be disabled by default, because I haven't seen it work at all. I made sure in viewport that I enabled "view culling/occlusion" while in wireframe and don't see any culling anywhere.

Has anyone else figured out how to get this feature to work properly? I have no intention of releasing my game for mobile platforms, so I do not wish to use pre-computed visibility.

[Graylord]

It does work, and is on by default.

[ankangronto]

while in wireframe mode it wont cull anything

[DIFTOW]

It does work, and is on by default.

while in wireframe mode it wont cull anything

If it is disabled in wireframe, then how are you supposed to see it working? Being able to see the culling occur is pretty important in making sure the culling is actually occuring properly.

What is the point of this option which I have selected?

[Crusha K. Rool]

I remember an option somewhere in the editor that would show in 2D viewports which objects were rendered in the active viewport and which not. There is also a console command to stop updating the rendering/occlusion calculations at a point while allowing you to continue moving around the level and see everything rendered that was visible from the point where you started the command.

[DIFTOW]

Is there a way to see this culling in perspective in a wireframe mode?

[Graylord]

I remember that being done in one of the 3d buzz tutorial videos, alas I have no idea how now.

[Spoondog]

Yea once you tick "view culling/occlusion" in perspective viewport, it should then show the culling happening in the orthographic viewports dynamically. There is no way to see it happening in perspective wireframe that I'm aware of, unless that's a new feature. However, in perspective in standard lit mode, you can type "freezerendering", and then fly around and you can see exactly what was rendered at the time, so you can see how well it's working that way.

The GPU occlusion is pretty fast really, especially if you're doing PC. But always give everything a CullDistance anyway if possible. I've never used the precomputed vis but I've heard the results aren't fantastic and it takes a lot of memory.

[DIFTOW]

Yea once you tick "view culling/occlusion" in perspective viewport, it should then show the culling happening in the orthographic viewports dynamically. There is no way to see it happening in perspective wireframe that I'm aware of, unless that's a new feature. However, in perspective in standard lit mode, you can type "freezerendering", and then fly around and you can see exactly what was rendered at the time, so you can see how well it's working that way.

The GPU occlusion is pretty fast really, especially if you're doing PC. But always give everything a CullDistance anyway if possible. I've never used the precomputed vis but I've heard the results aren't fantastic and it takes a lot of memory.

Thank you for the response. It does work with the 2D/orthographic displays, and if it is an accurate representation of the dynamic occlusion, it's not working very well on my landscape. It only removed one of the 4 components that make up my landscape, when one of the components is entirely out of view.

Is there anyway to divide a component into more pieces or make it cull per quad?

[Spoondog]

Actually, I remember with terrain, the culling in the editor didn't represent the actual culling that happened in-game, so it may be the same with landscape (e.g. meshes occluded by terrain were occluded in-game but not in the editor culling preview). I would go to that spot in-game and make sure it's working as you expect with freezerendering (remember it uses the bounds of the component when checking visibility).