RE: [Algorithms] PRT for noobs

[Peter-Pike S. <pp...@wi...>]

For "non-infinite" lights, I think there are two practical solutions:
=20
"Spherical Harmonic Gradients for Mid-Range Illumination", by Annen et =
al (http://graphics.csail.mit.edu/%7Ekautz/publications/index.html), I =
think using this where irradiance volume-like stuff is currently used =
makes the most sense.  This interacts very well with the various =
compression techniques as well.
=20
If you want a point/sphere light moving around a scene (or through an =
object or whatever), you can actually do this fairly easily (for direct =
lighting).  In general rotation of SH requires 3 DOF, however if the =
function has circular symmetry (for example a =
point/directional/spherical light), one of those DOF is redundant (ie: =
rotation around direction pointing to light), so you only need 2 DOF.  =
This means you can do the rotation using a (low res) cube map.  If you =
are using PCA to compress your PRT data, the cube map(s) just needs to =
store the projection of the rotated light into the PCA basis.  These =
"truly" local lights would require some math in the shader per-light, =
but for direct illumination this would essentially have zero error =
(indirect illumination will always be wrong for non-infinite lighting =
environments when doing precomputation...)  Of course all of the lights =
in the scene don't have to be handled with the same technique.
=20
For non-rigid objects, there are a couple of possible techniques:
=20
Skin the lighting environments (ie: rotate the distant lighting by each =
bone.)
=20
We have some new stuff ("Local Deformable PRT"), that the meltdown =
slides cover a bit that use a more approximate version of transfer which =
can be trivially rotated (we haven't submitted a publication on this yet =
and the more dynamic aspects are still being researched - in a nutshell =
we use zonal harmonics to approximate the transfer vector, and the =
convolution theorem to do the integration against the lighting =
environment.)  This could also be used as a normal map replacement (ie: =
no unique parameterization required) that handles general lighting and =
can model soft shadows and subsurface scattering.
=20
The two techniques above as stated are really only appropriate for =
"local" global illumination (ie: think of how bumps on a surface =
interact, not how limbs cast shadows on a body), but some general things =
(armpits are generally dark, etc.) are handled in a somewhat reasonable =
fashion.  Handling the "coarser" global illumination is an ongoing =
research problem.
=20
I think for more "scene centric"
=20
I would also recommend looking at the slides for our various talks, I =
think they are easier to follow than our papers.
=20
-Peter-Pike Sloan
=20

________________________________

From: gda...@li... on behalf of Aras =
Pranckevicius
Sent: Thu 9/30/2004 5:58 AM
To: gda...@li...
Subject: Re: [Algorithms] PRT for noobs




> Also, is this
> technique useful as a "general" solution to light your entire 3D
> static scenery? Or is it employed only for separated objects in a =
scene?

Not there yet... As you said, it's great for "environment" (light =
probes,
distant lights) lit separate objects, but not that good when lights are
close. And it doesn't work when lights are inside the convex hull of =
some
object (think a room-as-a-single-object, and light inside it).

Some other problems:
* non-rigid objects.
* light transport between objects.

Of course, the problems are in active research stage, but at the moment =
PRT
is not a general lighting (and light transport) solution.


Aras Pranckevicius aka NeARAZ
http://www.gim.ktu.lt/nesnausk/nearaz/



-------------------------------------------------------
This SF.net email is sponsored by: IT Product Guide on ITManagersJournal
Use IT products in your business? Tell us what you think of them. Give =
us
Your Opinions, Get Free ThinkGeek Gift Certificates! Click to find out =
more
http://productguide.itmanagersjournal.com/guidepromo.tmpl
_______________________________________________
GDAlgorithms-list mailing list
GDA...@li...
https://lists.sourceforge.net/lists/listinfo/gdalgorithms-list
Archives:
http://sourceforge.net/mailarchive/forum.php?forum_id=3D6188