Robert McGregor schrieb:
> I've been playing with the SSLT stuff the last 2 days and I noticed that in the
> Jensen SSLT SIGGRAPH paper there's a third RGB Diffuse Reflectance parameter
> (like a color_map for highlights, yes?) used to fine tune the material.

Which of the two papers are you referring to here? The 2001 
Jensen/Marschner/Levoy/Hanrahan paper ("A Practical Model for Subsurface 
Light Transport"), or the 2002(?) Jensen/Buhler follow-up ("A Rapid 
Hierarchical Rendering Technique for Translucent Materials")?

The 2001 paper had no such parameter: The scattering and absorption 
coefficients and the refractive index are essentially the only 
parameters in the model presented there. (I'm deliberately ignoring the 
phase function, as Jensen et al. themselves presumed isotropic 
scattering when they went on to measure the parameters of real-world 
materials.)

The 2002 paper does indeed mention a "diffuse reflection coefficient" 
with reference to the 2001 paper, Rd, but in the 2001 paper that was 
actually the /result/ of the SSLT computations (or, rather, with the 
formula referred to in the 2002 paper, even just an approximation).

The 2002 paper does indeed take this diffuse reflection coefficient as a 
parameter, but not in addition to the scattering and absorption 
coefficients appearing in the BSSRDF formula, but to reparameterize the 
whole smash, in order to compute the scattering and absorption 
coefficients from parameters that are more intuitive (said diffuse 
reflection coefficient, as well as the "mean free path" per color 
component).

The diffuse reflection coefficient, in that reparameterization, would be 
equivalent to the product of the pigment color and POV-Ray's the 
"diffuse" parameter.


 > So that
> would just mean extending the existing diffuse component of a finish to use a
> full RGB color vector instead of an assumed grayscale color vector
> as it is currently, which makes a lot of sense to me (and a second vector for
> the backlit stuff that Clipka did).

The diffuse colorization is what is currently in the pigment (which, as 
you will certainly agree, is much more flexible than just a color 
vector). Note that the pigment does not affect highlights or reflections 
unless used with the "metallic" keyword. Even then, they can be 
"decoupled" by using multi-layered textures.


> Regardless, accurate simulation of different physical materials requires
> different mathematical models to handle the various cases. I know what (and
> who) Phong is, but I'm not really sure what model POV-Ray's "specular"
> represents (just a guess - Cook-Torrence?).

No, it's actually just the Blinn-Phong model.

 > So, why limit POV-Ray to "phong"
> and "specular?" Why not just take into account several shading models for
> various materials, just like the high-dollar Hollywood boys do? (and they're
> pretty damned convincing most of the time)

Maybe because someone needs to implement the whole stuff? And maybe the 
people using it are not high-dollar Hollywood boys who know exactly when 
to employ what model?

And last not least, maybe it's also because the syntax to this day is 
not particularly inviting to add more alternative highlight models.

> That simply means making available various *combinable* shading models for
> various materials. Blinn/Phong for plastics, Lambert for simple
> non-reflective surfaces, Cook-Torrance for metals, Oren Nayar for rough
> surfaces, and Ward-anisotropic for objects with anisotropic reflections like
> brushed metal, fur, hair, etc. That's the utmost in flexibility as far as I can
> see.

I guess I do agree that it would be nice to have additional highlighting 
models, and maybe even be able to combine them.

However, given that the average user will use just /one/ of these many 
models, I would consider it a waste of memory (and a bit of computing 
time, too) to combine all of these side by side in each and every single 
texture finish.

I would therefore rather suggest to have a texture finish support 
exactly /one/ highlight model (let the user pick which one), which 
experienced users can then combine by using multi-layered textures.

Post a reply to this message

clipka <ano### [at] anonymousorg> wrote:
>
> The 2002 paper does indeed mention a "diffuse reflection coefficient"
> with reference to the 2001 paper, Rd, but in the 2001 paper that was
> actually the /result/ of the SSLT computations (or, rather, with the
> formula referred to in the 2002 paper, even just an approximation).
>
> The diffuse reflection coefficient, in that reparameterization, would be
> equivalent to the product of the pigment color and POV-Ray's the
> "diffuse" parameter.

Okay, I misunderstood completely then (which isn't terribly unusual). Thanks for
clarifying that.

> > So, why limit POV-Ray to "phong"
> > and "specular?" Why not just take into account several shading models for
> > various materials, just like the high-dollar Hollywood boys do? (and they're
> > pretty damned convincing most of the time)
>
> Maybe because someone needs to implement the whole stuff? And maybe the
> people using it are not high-dollar Hollywood boys who know exactly when
> to employ what model?

You are *so* warm and fuzzy sometimes :)

> And last not least, maybe it's also because the syntax to this day is
> not particularly inviting to add more alternative highlight models.
>
> > That simply means making available various *combinable* shading models for
> > various materials. Blinn/Phong for plastics, Lambert for simple
> > non-reflective surfaces, Cook-Torrance for metals, Oren Nayar for rough
> > surfaces, and Ward-anisotropic for objects with anisotropic reflections like
> > brushed metal, fur, hair, etc. That's the utmost in flexibility as far as I can
> > see.
>
> I guess I do agree that it would be nice to have additional highlighting
> models, and maybe even be able to combine them.
>
> However, given that the average user will use just /one/ of these many
> models, I would consider it a waste of memory (and a bit of computing
> time, too) to combine all of these side by side in each and every single
> texture finish.

As computing power continues to increase I don't think this is really an issue.
Besides, we're *ray-tracing*, which typically uses "a bit of computing time"
anyway, so let's make the most of it.

> I would therefore rather suggest to have a texture finish support
> exactly /one/ highlight model (let the user pick which one), which
> experienced users can then combine by using multi-layered textures.

Okay, I understand that. Although I often combine phong and specular in the same
finish block to get a wide, soft, general hightlight and a sharp specular
highlight it's no big deal to layer them instead.

I just wanted to prompt some thinking about adding some additional models (maybe
for POV-Ray 4). As far as syntax, I wouldn't suggest to change it, just add a
few keywords and maybe change some things under the hood. For example:

// use default Blinn-Phong
finish {specular 0.5}

// switch to Cook-Torrance
finish {specular 0.5 metallic}

// switch to Ward-anisotropic (for say, brushed metals)
finish {specular 0.5 anisotropic{gradient radial ramp {rgb 0, rgb 1}}}

Post a reply to this message

Robert McGregor wrote:

> Okay, I understand that. Although I often combine phong and specular in the same
> finish block to get a wide, soft, general hightlight and a sharp specular
> highlight it's no big deal to layer them instead.

But still sounds like we have a valid use case for using two highlight
models simultaneously, namely to approximate a third unsupported one. Of
course, that would no longer be necessary if the new syntax would some
day allow to specify a custom highlight model (as a function?), after
which the old keywords could be phased out.

Post a reply to this message

Zeger Knaepen wrote:

> I was thinking about that too, but it's inconsistent with current behaviour: 
> normally, specifying a value overwrites the previous value, which is a good 
> thing and imho must be kept.

arguably, adding highlights in a finish block could be seen
similarly to adding media to an interior or densities to media,
both cases where the effects are combined.

Post a reply to this message