Looks nice, what speed are you getting?

.edit: You say you're computing the average light direction... why not compute an average color for every ray that hits an object, and use that as the light's color? I think that'll give some nice looking radiosity effects. You may want to define a single (average) color per object as exact color computation would be too timeconsuming. Or maybe a small cube map instead of a single average color to produce more accurate results.

Or are you doing something like that already?

nice render, though i'd like to know: exactly what is it that you do once you have the average incident radiance direction?

moreover, when you say 'soft real-time lighting', do you mean quickly recalculating the lighting or just re-rendering from another viewpoint?

[pet peeve: 'radiosity' (being a crappy, discretised, diffuse-only* gi solution) is *NOT* the same thing as 'global illumination'. neither is radiosity an 'effect', there's actual maths and physics to this, and when people just... oh nevermind, no one cares anyway...]


* unless you use clustering or higher order approximations, which are a huge waste of time, code, blood, sweat, tears and most of all skill, when compared to elegant monte carlo methods...

>> [pet peeve: 'radiosity' (being a crappy, discretised, diffuse-only* gi solution) is *NOT* the same thing as 'global illumination'. neither is radiosity an 'effect', there's actual maths and physics to this, and when people just... oh nevermind, no one cares anyway...]

I agree, but my point was that what I said basically does what radiosity does, only on a much smaller scale and with a single iteration, namely shooting rays from discrete points on the surface to collect how much light (and what color) reaches that particular point.

Hi,

I precalculate the average direction of unoccluded rays and the average "intensity" (NumUnoccludedRays / NumRays). This is stored in a texture; the vector in the RGB channels and the intensity in the ALPHA channel.

The pixel shader just perform a texture read and calculate the diffuse light using lambert's cosine law multiplied by the intensity. DiffuseLight = dot(AvgVector, normalize(LightPos - FragmentPos)) * Intensity


/ Sunray

"shooting rays from discrete points on the surface..."

*ahem*, this isn't what discretisation is, and what you described isn't how radiosity works.

hmm, people often see corrections as being an attack (*sigh*), so i'll leave it there...

> DiffuseLight = dot(AvgVector, normalize(LightPos - FragmentPos)) * Intensity

so you're using the average incident light direction as the normal, then multiplying by the occlusion factor... need to work out what this is doing, and also why ;)

gonna get your program and try see what happens when the light moves... not quite sure how ambient occlusion ties in with that :)

hmm, now that i *really* think about it, excessive nitpicking is just stupid.

apologies ^_^