I am planning to use something like irradiance environment-maps for the ambient lighting in my engine. Whether that''d be good or not, i don''t know. But i think it would look cool in outdoor scenes. And it is relatively cheap compared to radiosity .

It is only a texture lookup. As texcoord the normal in worldspace is used. And the irradiance environmap would be a downsampled and blurred version of the environment map.

Sounds cool, doesn''t it? But i would like to hear if this could replace the constant ambient term.

OK thanks Yann, I will give it a try! If floating-point buffers are not available, I can still fall back to standard A8R8G8B8.

No, I think you''re right - well placed diffuse light sources do a better job than a simple ambient term.

However, setting up a simple ambient term is much easier than setting up "well placed" diffuse lights.

Also, if the diffuse lights cast shadows then it kinda ruins the effect btw.

Remember that website on radiosity (hugo elias'')
http://freespace.virgin.net/hugo.elias/radiosity/radiosity.htm

He shows his example of the difference between radiosity and no radiosity.
Notice in the "no radiosity" example he actually adds a directional light so you can at least see the geometry instead of just black with a little red.
If he had added just an ambient term it would have looked solid dark red (no way to see the pillar''s edges against the walls) and solid dark gray for the walls.

A directional light does wonders if you don''t want a very dark scene area. So yeah, in cases where you''d use an ambient term, a single directional light (not very bright) would probably look better.

You might think about a directional light source (no shadow calculation) instead of many point lights (cheaper calculations too) because it would be much easier on the artists.

-Michael

Hugo Elias' Radiosity Tutorial. Must Read!

Cows:
Realtime VertexRad is easy enough to achieve, provided you don't mind sticking with the simplest geometry - meaning a square room is simply a vertex in each corner. 8 vertices per room is dead easy to get realtime, but it looks awful 99% of the time. To get anything remotely good looking, you need to tessellate your geometry to the same sort of level as you would if you were lightmapping. Not only does this then bring the calculation time into line with a lightmapped solution, it also increases the vertex count, thus stressing the GPU more. Of course, once the VRad has been calculated, you can certainly store the resulting color values to be loaded at run time.

My conclusion is that VRad isn't going to be useful until CPU and GPU speeds increase significantly, along with the data rate between the two (for transferring huge quantities of vertex data). By this time, lightmapping will be obselete (for the most part), and someone (probably YannL ) will have come up with a shiny new lighting system to wow us all with.

EDIT: damn clicky! when am i ever gonna learn?!


[edited by - blacksheep on September 4, 2003 4:57:02 AM]