Unity3D : Performance with multi-camera rendering with a lot of spheric meshes

Question

I have an Unity 3D scene with several cameras looking at the same object (a huge brain mesh ~100k tri) but not necessary with the same point of view.

In the same 3D scene there is a huge number of spheric plots meshes (from 100 to 30000).

In all the cameras i have to display the brain mesh with a part of the plots meshes.

Depending on the camera view, each plot can have a different size (mesh filter and spheric collider), a different material (opaque or transparent) and can be visible or not. The spheric collider must have the same size than the mesh.

I set up a shared mesh in common for each spheric mesh. Their material can be one of the several shared materials i have defined.

Before rendering the scene, for each camera view in the OnPreCull function i have to define which plots are visibles and how they look.

This part can be very costly, i tried several things :

• setting gameobject inactive : too costly
• setting local scale to vector3(0,0,0) : better but i can see that the rendering is still done in the profiler
• setting a total transparent material : same result, but the in the profiler the rendering is now transparent instead of opaque
• setting a layer not in the cameras layers masks : huge script cost

I don't kwnow if i can make an efficient culling system with all theses cameras looking at the same point...

I welcome any new ideas.

Answer 1

First issue:

Regarding your specific question with the four dots.

Simply set the renderer.enabled = false, that's all there is to it.

Note however that as I mention in a comment, you would never try to "cull yourself" in Unity (unless I have misunderstood your description).

---

Second issue:

Regarding the small spheres. I suspect you have very many in the scene. You simply can't do that. In video games (the most difficult of all 3D engineering), you just do this with billboarding. It's how say "grass" is done in a scene. You can achieve this nicely with the particle system in Unity, or other techniques. An implementation is beyond the scope of this answer, but you will have to fully investigate billboarding. Simply it's a small flat image which always faces the camera in the render pass.

---

Issue 2B:

Note however that sphere colliders are wonderful, and you can use as many as you want. I'm sure this is obvious from base mathematical reasons. Side tip: often folks try to "write their own" thinking it will be faster. It's impossible to outwrite the 100? person-years of spatial culling scientific research in PhysX, and moreover they use the metal, the gpu, so you can't beat it.

---

Issue three:

Is there a chance you're using a mesh collider somewhere in the project? Never use mesh colliders, at all. (It's extremely confusing they are mentioned or used in Unity; they only have one or two very specific limited uses.)

---

Issue four:

I'm confused about why you are turning things on and off. I have a guess.

I suspect you are not using more than one "stage"!

There's an amazing trick about video games when you have more than one camera. In fact you have "offscreen" scenes! So you may have players in a dungeon or the like. Off "to the side" you may have an entirely duplicate or triplicate setup of the whole thing running (you could "see it if the camera turned the wrong way") for the other cameras. (In the example you would have different qualities on the dopplegangers, coloring, map-style or whatever the case is.) Sometimes you make a whole double just to run physics calculations or address other problems.

Fascinating extreme example of that sort of thing.

In short in your situation,

You likely need one whole 'stage' of a camera and brain for each of the camera views!

Again this can be http://answers.unity3d.com/answers/299823/view.html but it is indeed the everyday thing. In your overall scene you will see eight happy brains sitting in a row, each with their own camera. In each one you would display whatever items/angle etc are relevant. (Obviously, if certain items are "identical, other than the viewing angle" you could use the "same brain with more than one camera": but I would not do that, best to have one-brain-one-camera for each view.)

I believe that could be the fundamental issue you're having!