Global Illumination creator for Static Meshes

Mel · Post by **Mel** » Thu Jun 02, 2011 4:30 pm

devsh wrote: ...The handle in this case is a u32 index (number)...

i would suggest that you used, if posible, void* pointers, because using u32 won't be portable to 64 bit compilers, which is something that in the world of GI calculation you really want.

devsh · Post by **devsh** » Thu Jun 02, 2011 8:24 pm

looking into icospheres for equidistant points...

LizardGamer · Post by **LizardGamer** » Fri Jun 03, 2011 1:00 am

mongoose7 wrote:
devsh wrote:Anyone knows a good lossless compression system for images??? (I could use zip but its really not made for the job)
Maybe Google "huffyuv"?

Ha-ha, equidistant points on a sphere. Probably equivalent to the Riemann Hypothesis.

huffyuv is a video compression and I'm sure it only works on windows

mongoose7 · Post by **mongoose7** » Fri Jun 03, 2011 3:40 am

Yes, but the compression is intraframe, so it could work on images. However, if there is no source then that is moot.

devsh · Post by **devsh** » Sat Jun 04, 2011 8:55 pm

I'm hoping to get 10x speed increase from doing the probe calculation on the GPU

devsh · Post by **devsh** » Mon Jun 13, 2011 6:36 pm

dues to the very unfortunate construction of the complete GI approach, I have to construct an entire renderer.... currently implementing inferred lighting in irrlicht.

So that means I gotta lay off this project until that is done.

baiqian · Post by **baiqian** » Tue Jun 14, 2011 8:00 am

This static ambient occlusion is impressive based on TV3D, it's a good solution for GI. I really want to try move it to irrlicht, but I'm not quite understand the algorithm or VB.net(the sample). It use both raytracer and hemicube two solution to calculate and save the info into the model. And on my HD 3650, it runs 600+FPS.
http://theinstructionlimit.com/static-ambient-occlusion
http://www.truevision3d.com/forums/show ... 674.0.html[/img]

Radikalizm · Post by **Radikalizm** » Tue Jun 14, 2011 10:06 am

baiqian wrote:This static ambient occlusion is impressive based on TV3D, it's a good solution for GI. I really want to try move it to irrlicht, but I'm not quite understand the algorithm or VB.net(the sample). It use both raytracer and hemicube two solution to calculate and save the info into the model. And on my HD 3650, it runs 600+FPS.
http://theinstructionlimit.com/static-ambient-occlusion
http://www.truevision3d.com/forums/show ... 674.0.html[/img]

The huge downside to this technique is that it's completely static and therefore not suitable for dynamic environments (ie. you can't move any objects or the lighting will appear incorrect)
Also, this technique is per-vertex so I doubt it'd look that impressive on meshes with a low vertex density
And devsh is talking about a Global Illumination algorithm, not just an ambient occlusion algo; it's true that AO plays its role in GI calculations, but AO is not the same as a full GI implementation

devsh · Post by **devsh** » Thu Jun 16, 2011 6:21 pm

haha.. from what I gathered from the truvision3d forum.... the GI is made by iterating through the triangles of the model and counting how much each contributes to GI

Obviously this fastly becomes a O(n!) calculation which has similar performance to a brute force n-body simulation, not mentioning you have to do it for every frame of an animation and skinned meshes totally F***ck up

In fact... if you really wanted to I could write you a global illumination function in C++ which would calculate disk to disk based AO+GI based on the Nvidia paper in under 3 hours.

devsh · Post by **devsh** » Sat Jun 18, 2011 11:52 am

For my inferred lighting I have a 128bit gbuffer, 80bit scalegbuffer, 128bit lbuffer at half the resolution, and a 64 bit framebuffer.... all this takes up a mere 32mb at 720p (with a medium quality setting)