"gregjohn" <pte### [at] yahoocom> wrote:
> Here are some new experiments based on a better following of advice of Christoph
> and the SIGGRAPH paper:
> http://www-graphics.stanford.edu/papers/bssrdf/bssrdf.pdf
>
> I used their Skin1 (I presume African), Skin2 (Caucasian/ Asain), and Chicken1
> (completely unexpected-- even for "dark meat".).  Compared are my previous
> "mistake" of basing the subsurface parameters on a typical skin color, and using
> diffusion.  And that's all compared to a typical pre-SSS texturing.  The image
> was labeled in Powerpoint and screen-grabbed from there, but I don't think the
> quality is adversely affected.
>
> Comments:
> i) The non-SSS looks so hideous I don't know if I want to render a human that
> way again.
> ii) The paper talks about Fresnel reflection, but these people are a bit too
> reflective. And um, I don't know what specular to use.
> iii) All the SSS folks have unusually dark fingers and ears. This may be the
> limitation of the technique.  Lower values of mm_per_unit give darker fingers;
> higher values make some of the vertices stand out in an unusual way, one which
> is just starting to happen in this image.
> iv) With any future experimentation I think I would start adding some diffuse. I
> bet that'll fix the dark ears & fingers and make it a little less waxy.
>
> PS. I could "donate" anonymously, CC-Zero license, the mesh exported from this
> Circa 2009 version of makehuman, if "povray" wants it.

Very useful experiment. Here are some comments:

-The dark blueish tip of the fingers is something I've met too in my tests. In
other software it's rather the opposite: brighter and more towards the red
extremities. Though it may be caused by a wrong scale for the sss phenomenon, it
shows that the SSS is currently hard to handle, not flexible/forgiving enough,
otherwise, if the mm_per_unit scale is right, maybe it shows that there is some
error somewhere in the current POV implementation?

-for specular or reflection, what is important is that the human body has
different levels of shininess and roughness per each area. so it needs to be
texture mapped, which is a pain (but possible) with the current syntax.
specular_map would be a delight here and the specularity should be as broad as
possible, if using fresnel reflection instead we would then need blurry
reflections which is also a pain (but possible) with the current syntax.
The best you can do for now (so you don't always need to use diffuse) is to use
a much rougher specular and probably a very low value.

-About the dark chicken, it's not a pov mistake, this sss preset also turns out
that dark in Blender. The only way a chicken might look like that is without the
skin and before cooking so maybe CG artists could use this preset as a basis for
raw meat. I suspect it could also be a private joke of the guys who made the
paper, maybe a chocolate chicken... :-)

-About the shiny vertices in the diffuse+sss version, I would rather say they
are accurate as you seem to have used a non smoothed mesh. you can smooth it in
blender 2.5 and export to POV from there now. You'd just need to find the
temporary folder where the blender file is exported. to copy and paste mesh2
data in yours.

-If you made more tests, maybe with all the values of the paper, it would be
nice to have the values for each keyword, and maybe render times?

Post a reply to this message