Case of the vanishing meshes: culling?

[Rastar]

I am laying out a large array of mesh patches that will be heightmap displaced in the shader. It's working fine, except that when the camera moves a certain distance from the origin (about 1500 units or so) all meshes suddenly vanish. Moving a little back they appear again. This is independent of the camera's clipping planes.

 

Bug or feature? If feature: Is there a way around this?

 

There...

 

...and gone. The meshes in the back are gone as well.

[shadmar]

Sounds to me that AABB for your meshes is not updated.

[Rastar]

I call

 

patch->UpdateAABB(Entity::LocalAABB|Entity::GlobalAABB);

 

after creating and/or scaling a patch. But a question: I am displacing many of those patches a fair amaount after creating them. Do I have to call UpdateAABB(Entity::GlobalAABB) after moving them?

[shadmar]

http://www.leadwerks.com/werkspace/page/documentation/_/command-reference/surface/surfaceupdate-r638 perhaps?

[Rastar]

Mmmh, no, both Update() nethods don't seem to help.

[shadmar]

Try diable culling or set aabb manually. I dont know, usually it's aabb related.

[Flexman]

As Shadmar says, culling is done using AABB. Workarounds might include manually adjusting the AABB values (then calling the aabb update) or adding a couple of vertices to the mesh to make it bigger.

 

Could be a bug though, had many frustrating times with objects being culled. Are these patches parented at all?

[Rastar]

No, currently they are not parented. Do they have to be?

 

I have a set of 7 different patch types (Models) which I instance, scale and position to create the "mesh puzzle" seen in the screenshot.

[Josh]

It's hard to say without seeing your code.

 

If the local AABB is correct then the engine will automatically adjust the global and recursive AABBs whenever things move. There is also a cull distance command that will make them disappear a certain distance away.

[Rastar]

Well, here is my code (don't be too harsh, I'm in the process of relearning C++ ... ;-)

 

I am trying an implementation of the geometric clipmap method as described here http://http.developer.nvidia.com/GPUGems2/gpugems2_chapter02.html. It uses a series of concentric mesh rings with decreasing LOD, each ring assembled from several patches.

 

As described, run forward using "W" to about 1500 units and all patches vanish.

 

Thanks!

[Josh]

Try adding this to the end of the RectangleTesselator::Computer function:

surface->UpdateAABB();

 

I am curious why you are duplicating the built-in terrain. That paper is describing almost the same exact technique I implemented. (Their geometry approach is a little different, but it looks like the texturing technique is identical.) Not that I mind.

 

Their geometry approach is nice in that it would only require a few passes, although it would be more restrictive with LOD distances and max view range. Something to think about at some later point...

[Rastar]

Whoaa, I wasn't aware of that. I noticed your texturing approach looked similar to their structure for the heightmap base mesh. I am heading for a very large terrain, and their approach seemed very interesting to me, especially with the method for compressing the height data. If I can use that with the built-in terrain I would be more than happy!

 

I've tried the various UpdateAABB() calls, but that doesn't seem to help.

[Josh]

Well you're missing the Surface::UpdateAABB call right now, and it is required:

http://www.leadwerks.com/werkspace/page/documentation/_/command-reference/model/modelcone-r348

[Rastar]

I understand, but adding that call to the RectangleTesselator doesn't change the behavior, at least on my end.

 

Just wanted To say: I really appreciate your support, I know you have a load of stuff on your plate and more urgent business to do than fixing other people's code...