PDA

View Full Version : HW Global Illu screenshot



McBain
03-10-2002, 02:54 PM
I am proud to present,
the very first thing coming out of my global illu algo on the gf3

Its only a rough approximation, and i'm implementing a dozen things to improve it now

but its already nice

notice for example the indirect lightning on the ceiling (it would be black with only direct lightning), the soft shadows coming from the area light source, and the fact that shadows are not 100% black (also due to indirect lightning).

anyway, here it is:
http://users.pandora.be/lagae/indirect/ss0020.jpg

Diapolo
03-10-2002, 03:56 PM
Well, I dunno how to do such things (perhaps IŽll learn it sometime? *g*), BUT ... it looks really sweet, keep up that great work and post links to new screenies / demos http://www.opengl.org/discussion_boards/ubb/smile.gif.

Diapolo

davepermen
03-10-2002, 10:25 PM
looks amazing.. any infos how done?

kon
03-10-2002, 11:21 PM
Looks good, but only 1.57 fps? What are you calculating? Radiosity?

kon

Tom Nuydens
03-11-2002, 12:06 AM
Originally posted by kon:
Looks good, but only 1.57 fps? What are you calculating? Radiosity?

kon

He's got indirect lighting! I don't know if it's radiosity or something else, but 1.6 fps is bloody impressive http://www.opengl.org/discussion_boards/ubb/smile.gif

-- Tom

richardve
03-11-2002, 12:22 AM
[friendly hugmode]

Amazing!

Would you be able to use some more colors and upload some more screenshots?

Bye!

[/friendly hugmode]

T2k
03-11-2002, 01:02 AM
nice, but try to optimize(try to cache data if its not changed !!), you can push your framerate up to more than i can see there, except you have formed the scene with more than ~100Tris

Carmacksutra
03-11-2002, 07:32 AM
Interesting.

Does it work with curved surfaces ?
(I mean: do curved/tesselated surfaces slow it down a lot, compared to large flat polygons?)

McBain
03-11-2002, 08:04 AM
Originally posted by Carmacksutra:
Interesting.

Does it work with curved surfaces ?
(I mean: do curved/tesselated surfaces slow it down a lot, compared to large flat polygons?)

it works with everything http://www.opengl.org/discussion_boards/ubb/smile.gif
but my scene management does tesselate curved surfaces correct for the moment ...
and yes, offcourse it will slow down

davepermen
03-11-2002, 08:12 AM
i guess because its hardware accelerated he uses render to texture or some other funny thing.. but yes i think its always dependend on the amount of triangles he's using http://www.opengl.org/discussion_boards/ubb/wink.gif

btw, there are much nice pics on your page, really.. the different brdf's, the different shadowing-approaches etc.. very cool.

would be nice to get some infos from where you have your knowledge or how you could share it with us..

McBain
03-11-2002, 10:27 AM
i've put some color in it http://users.pandora.be/lagae/indirect/ss0025.jpg

what disturbs me the most in this image is it is to dark (eg the front of the yellow cube). This is because the box is open and light escapes, never getting reflected by something else to get back to eg the yellow cube.

Interpreting your results is often more difficult than implementing the algorithm.

from where i got the knowledge ?
I'm a masters student at a the computer graphics research group @ KUL university , leuven, belgium. So, i have plkenty of people who can give pointers to usefull material.

It is important, that techniques are best learned in theory first. First learn eg how to do shadow mapping, don't learn how to do shadow mapping *in opengl*. Once you learned the technique, it can (sometimes awkward) be implemented on most gfx api.

Personally, i started with the book computer graphics of Foley. A very interesting book. Not about opengl. "Computer graphics using opengl" hast, next to opengl learning code, also interesting topics. Also, when you decide to start with opengl, the redbook is a must. When you work with extensions, nvidia has some great papers. The most important up to date source of knowledge are defenitely the papers. (like SIGGRAPH papers). This is where, for example, engeneers @ NVIDIA get their knowlegde. The nvidia brdf for example is based on papers of Kautz and McCool and ...
A lot of new things get introduces in nvidia's site by recent papers Eg ASM (adaptive shadow maps), an interesting technique, but not yet implemented by nvidia engeneers (i think). I guess they will publish a paper on their site about it soon.
Also, dont focus anly on realtime gfx. Al lot of techniques of global illu, radiosity, raytracers, ... can be used in realtime gfx, and vice versa. Sometimes you can adapt these techniques or use elements of it in real time stuff.

anyway, being up to date is a full time job and no-one does know everything about everything.

T2k
03-11-2002, 10:25 PM
you have an GeForce3 PCI ???

McBain
03-12-2002, 02:15 AM
Guess u took a look at my other screenies http://www.opengl.org/discussion_boards/ubb/smile.gif
The GL_RENDERER string says so, but i doubt it very much http://www.opengl.org/discussion_boards/ubb/smile.gif

richardve
03-12-2002, 02:54 AM
I think you should install an AGP driver http://www.opengl.org/discussion_boards/ubb/smile.gif

TM
03-21-2002, 03:58 AM
Interesting pics...
Are u using instant radiosity? Do u use hw accelerated shadow maps? is this technique view-dependent?

tnx

SirKnight
03-21-2002, 05:45 PM
Would it be possible for us to get the exe? I would like to try it on my GeForce 4 Ti and see how it runs on there.

-SirKnight

zed
03-21-2002, 10:29 PM
ive been doing quite a bit of reasearch on this also in the past
(appear soon in a flipcode IOTD near you http://www.opengl.org/discussion_boards/ubb/smile.gif) http://uk.geocities.com/sloppyturds/rad.jpg
±10fps celeron433 gf2mx
+ no u cant have a demo cause its top secret
btw its choice as to drag a box around + watch the lighting change in realtime

knackered
03-22-2002, 04:16 AM
It seems to me that he's just doing a low number of passes with a normal radiosity algo, writing into the textures. On a powerful enough machine (above 800mhz) I'd expect those kind of frame rates.
What makes you all think he's using some special hardware features?

LordKronos
03-22-2002, 04:40 AM
Originally posted by zed:
no u cant have a demo cause its top secret


Appearantly, even the image that you posted a URL to is top secret, because geocities wont let me get to it http://www.opengl.org/discussion_boards/ubb/smile.gif

kon
03-22-2002, 05:06 AM
Just copy/paste the url and you'll see the image.

kon

Carmacksutra
03-22-2002, 06:18 AM
Originally posted by knackered:
It seems to me that he's just doing a low number of passes with a normal radiosity algo, writing into the textures. (...) What makes you all think he's using some special hardware features?

I doubt McBain's algo has much to do with "normal" radiosity algorithm (I don't know what Knackered exactly means to call "normal radiosity algorithm", but for me it is rendering from patch point of view)

On Zed's image there are visible linear-interpolation patterns, typical to well-known lightmaps (i can even guess that front facing wall is subdivided to about 8x7 patches). So I believe Zed uses something close to "normal" radiosity, but is working on speeding it up.

But this doesnt apply to McBain's image.
On bright walls there are visible multiple overlapped low-contrast but sharp-edged shadows. He must be using something different, not requiring subdivision to patches. (OTOH this makes caching of results a lot harder?)
Let me speculate (dont laugh http://www.opengl.org/discussion_boards/ubb/wink.gif): Is it kind of recursive projecting every light-receiving poly onto surrounding ones, with use of standard shadows ?

McBain, help us solve your puzzle http://www.opengl.org/discussion_boards/ubb/wink.gif

TM
03-22-2002, 11:03 PM
I think McBain uses some variant of the instant radiosity algorithm. This technique doesn't need patch subdivision nor light mapping, but instead computes radiosity in image-space.
You can find an intuitive description of the algorithm here: http://www.cs.utah.edu/~schmelze/radiosity/proj/proj.html

zed
03-23-2002, 01:51 AM
thanks for that link TM
there was a link off that http://www.uni-kl.de/AG-Heinrich/Alex.html
which seems like a good read, i must get busy http://www.opengl.org/discussion_boards/ubb/smile.gif