Re: [grt-talk] Some suggestions.

[Nikodemus Siivola]

On Wed, 30 Apr 2003, Simon Adameit wrote:

> Highlights are *blurred* reflection of light-sources. afaik phong is a
> cheap trick to simulate these using the two parameters roughness and
> reflectivity it is not related to light-source size. And I think it
> wouldn't be very usefull if we did this as light-sources big enough that
> this would be necessary should be done as area-lights anyway.
> So we better implement area-lights and let them not only generate area
> shadows but also area highlights and diffuse.

I was obiously too terse.

Yes, phong is a trick, controlled by two parameters. And yes, these are
traditionally *called* roughness and reflectivity, which really makes no
sense: a rough surface is not going to be reflective. It all comes down to
phong being a trick.

Area lights are a must. Eventually. But they are also computationally
expensive: in some cases it's quite beneficial to be able to just simulate
them with cheap tricks: for previews, and for images that would take too
long to render otherwise.

See:

  http://www.cs.utah.edu/~bes/papers/coneShadow/shadow.html

I want to eventually let all light-sources (both area and point) have a
looks-like attribute that becomes their reflection / highlight: for point
lights the bounding box of the looks-like object can be used to guess the
virtual size for the simulated soft shadows (or it can be manually
overridden). In case there is no looks-like a virtual size can still be
specified to enable the "smart phong" and simulated soft shadows.

Of course, classic phong model can / will be incorporated as well. And it
may be that this quest to empower the point-light's is misguided and doomed
to fail. I've resolved to try, however. ,)=

Smart phong would seem to have an advantage over the classic one: with
classic phong you have identical higlights for all lights. With smart phong
they can vary with distance and virtual size with no extra computational
expense. I'm hoping that this might prove to be a case of "any detail is
better than no detail".

About blurring: "raytraced look" is unnaturally sharp by it's basic nature.
For realism blurred reflections, shadows and focal blur are all needed,
along with diffuse interactions. For flexibility in image generation these
should be as orthogonal as possible. The funny thing is that they all come
down to cheap tricks vs. stochastic methods -- I think we should try to
accommodate both.

Your point is well taken, however. I will implement the classic model as
well, so that the effects can be compared side by side. I mean, how else
can the viability of the alternative technique be assessed? And if it sucks
we still have the classic phong...

Cheers,

  -- Nikodemus