fake radiosity

[Diego]

hi!

i was taking a look at this website: http://lightsprint.com/ (very cool stuff, btw) and started wandering about a possible method for fake radiosity for games. (not from a distant future)

the only thing i can think of is scattering (strategically) omni lights near surfaces (with the same color too), and make it 'light sensible'. by that i mean it should be brighter if lit by another light source.

if anyone is even more bored than i am to try do something like that, please show us! lol

[oDDity]

They already have more or less fake radiosity in the new crysis engine. I'm sure it won't be so long before we have 'real' radiosity in game engines, with actual ray tracing and photon bouncing. Near realtime renderers are already being developed for applications like architectural visualisation.

Global illumination/final gather/radiosity has rapidly been getting faster and faster in 3d renderers. With my quad core I can render GI scenes now in 3 minutes that would have taken an hour 3 years ago with an older machine.

Real time won't be far away.

However, for most games, I can't see how it's going to improve anything, other than looking a bit more pretty.

[Diego]

yep.. (distant future = 4 years or something)

and well, about improvements.. it's going to improve the same as normal maps, dynamical lights, etc: it will be prettier and we're gonna have to buy better hardware.

the good news is that by then i'll be ably to play the next gen games of now! lol

[Gildoran]

I've long had an idea for a way to do real-time normalmapped radiosity, though my idea doesn't reflect light off of portable objects. The idea is that you divide BSP space into a tetrahedral mesh, and use a precalculated radiosity matrix to quickly convert light at the surfaces into the correct radiosity at the tetrahedral mesh vertices (this could be done very fast by abusing the graphics card). Then you render the scene, filling the depth buffer, and render the tetrahedral mesh, linearly interpolating lighting between its vertices (I believe it's possible to do this on current generation graphics cards, though I could only figure out how to render one tetrahedron at a time, so that may be a bottle-neck). Rendering a single tetrahedron can be accomplished by using the stencil buffer to ignore any fragments outside of the tetrahedron, then pushing the position/depth of each fragment through an inverse screen matrix and a tetrahedron shape matrix to find out how much "weight" to give to each vertex's lighting.

 

Making the radiosity normalmapped is a simple matter of computing a low-res cubemap for each vertex instead of a single value.

 

Ambient occlusion for portable objects would be possible also using tetrahedral occlusion meshes that are attached to the objects.

 

I believe that for scenes with very simple BSP geometry, this could potentially be fast enough to run on current hardware.