Jump to content

  • Log In with Google      Sign In   
  • Create Account

Banner advertising on our site currently available from just $5!


1. Learn about the promo. 2. Sign up for GDNet+. 3. Set up your advert!


Culling out-of-frustum models that contribute shadows


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
6 replies to this topic

#1 JMab   Members   -  Reputation: 228

Like
0Likes
Like

Posted 02 September 2014 - 06:38 AM

Hello all,

 

Looking for some collected wisdom here on this best strategy for this issue.

 

I'm performing simple frustum culling of models, and also shadow mapping combos of point/spot lights and models. That all works fine. However, there is one problem I've just noticed. When I move the camera to a position in the scene where a shadow-casting model becomes frustum culled, even though it's shadow should still be seen within the frustum, the model is removed from the render queue and no shadow mapping is performed for that model.

 

I'm thinking that the naive general approach would be to maintain a 2nd list of render commands of models for shadow casting that have not been frustum culled (although the list would probably still be reduced by some other scene visibility approach). Or I could go for a specific approach and let the artists tag shadow casters for no-culling when the camera is within a user-defined zone.

 

Is there a clever general approach to this? Should I be calculating the volume of the lights (I can do this in shaders already for debugging), calculating what models intersect those volumes and stop them from being frustum culled? Is that the best general approach?

 

Thanks in advance.


Edited by JMab, 02 September 2014 - 06:39 AM.


Sponsor:

#2 TiagoCosta   Crossbones+   -  Reputation: 2769

Like
7Likes
Like

Posted 02 September 2014 - 07:31 AM

Is there a clever general approach to this? Should I be calculating the volume of the lights (I can do this in shaders already for debugging), calculating what models intersect those volumes and stop them from being frustum culled? Is that the best general approach?

 

You almost have the right idea.

 

When rendering shadow maps, you should perform frustum culling but instead of using the camera frustum, you should calculate a different frustum using the lights' view and projection matrices (that you use to draw the models to the shadow maps).

 

So when creating shadow maps you should do something like this:

foreach shadow_casting_light i
{
     generate frustum from light i view/projection matrices;
     
     test models against frustum;

     render visibles models to shadow map;
}

So if you have n shadow casting lights, you'll have n lists of visible models (one for each light) plus another list for the final rendering from the camera perspective.

 

For point lights, you can generate 6 lists for each light (one for each face of the cube shadow map) or simply test which models are within light bounding sphere, or both or something else.


Edited by TiagoCosta, 02 September 2014 - 07:36 AM.


#3 JMab   Members   -  Reputation: 228

Like
0Likes
Like

Posted 02 September 2014 - 02:13 PM

OK, thanks Tiago. I can see that is the perfect general approach.

 

A lot of frustum culling though! I've been following an "only implement when I need it approach" to this engine/game development, and had been getting by with frustum culling to the scene camera for the entire list of scene entities, however given that I'll need 6 additional lists per point light and 1 per spotlight, I'm going to have to implement spatial subdivision structure, maybe a loose octree, to minimize culling.

 

Thanks again!



#4 Hodgman   Moderators   -  Reputation: 37634

Like
2Likes
Like

Posted 02 September 2014 - 03:59 PM

Lots of engines seem to test all the frustra at once, storing the results in a bitmask per entity --
E.g.
For each model
  Model.visible = 0
  For each frustum, i=index
    If model in frustum
      Model.visible |= 1<<i
Unless you're doing a huge number, say 100k, of these tests, it should run fine without any acceleration structures :)

#5 JMab   Members   -  Reputation: 228

Like
0Likes
Like

Posted 02 September 2014 - 06:57 PM

OK, that makes sense - compile a list of all of my scene camera and shadow-generating light frusta and then use that list when traversing the scene. Store results in bit flags for querying later in the frame. Means you only have to traverse the scene once, which would be a slight performance improvement over traversing for each frustum.

 

I'll give it a go without an acceleration structure - good practice not to try and prematurely optimize!


Edited by JMab, 02 September 2014 - 06:59 PM.


#6 frob   Moderators   -  Reputation: 27967

Like
1Likes
Like

Posted 02 September 2014 - 07:15 PM

Just be careful about scale. These are nested loops in the form "for all light sources, for all models that can shadow, for all models that can be shadowed, do some potentially expensive computing." You're looking at exponentially growing numbers of tests that are not necessarily fast to perform.

If any of those can have a large numbers, lots of lights, lots of shadow casters, or lots of potentially shadowed objects, your simulation can hit nasty combinatorial explosions.

It is especially dangerous for oblique shadows, shadows very near a distant light source, or any other case where a shadow can potentially interact with a large number of potential scene objects without a trivial culling.

Check out my book, Game Development with Unity, aimed at beginners who want to build fun games fast.

Also check out my personal website at bryanwagstaff.com, where I write about assorted stuff.


#7 JMab   Members   -  Reputation: 228

Like
0Likes
Like

Posted 02 September 2014 - 08:01 PM

Sure, I think I'll need to reasonably aggressively cull shadow-generating lights - I've had that as a TODO in my code to do this for a while and am assuming that, that culling will be straightforward, given that my scenes will typically be heavy only on point lights where simple sphere/frustum intersections can discard non-scene-affecting lights quickly.

 

I'm allowing lights to be flagged as non-shadow-generating and models as non-shadow-casting and/or non-shadow-receiving, so there are those possibilities for performance optimization.

 

It does make me wonder whether I'll really get the benefit out of my deferred shading path though - it's all well and good to be able to light the scene with dozens to hundreds of non-shadowing-generating lights, but as soon as you implement shadowing, you're brought back down to Earth!






Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS