after reading HW Jensen bssrdf paper (in case it'll be easy to code :) i
have a question : one part of the algorithm requires a monte carlo
integration with points lying on the surface around a point x0. So given a
float value d, we need a random point xi lying on the surface at distance d
of x0
anyone have an idea how such points can be found in a ray-tracer (we only
have a trace function) ?

M

Post a reply to this message

Mael wrote:
> 
> after reading HW Jensen bssrdf paper (in case it'll be easy to code :) i
> have a question : one part of the algorithm requires a monte carlo
> integration with points lying on the surface around a point x0. So given a
> float value d, we need a random point xi lying on the surface at distance d
> of x0
> anyone have an idea how such points can be found in a ray-tracer (we only
> have a trace function) ?

I'm not much into Jensen's paper (although i read it some time ago) but
isn't it possible to assume the surface is locally flat?

I just had a quick look again and i think this is generally assumed
(although it's mentioned that for highly curved surfaces special
consideration is necessary)

Christoph

-- 
Christoph Hormann <chr### [at] gmxde>
IsoWood include, radiosity tutorial, TransSkin and other 
things on: http://www.schunter.etc.tu-bs.de/~chris/

Post a reply to this message

> I'm not much into Jensen's paper (although i read it some time ago) but
> isn't it possible to assume the surface is locally flat?

hmm yes , but i m pretty sure bad situations will arise, for points near the
edge of a cube for instance

M

Post a reply to this message

Mael wrote:
> 
> hmm yes , but i m pretty sure bad situations will arise, for points near the
> edge of a cube for instance
> 

As i understood the paper, these problems can be avoided by always using
the same distance for calculating the diffusion term instead of using the
individual distance of each sample.  It also says this is likely to
produce some inaccuracy but it seems to work quite well.

Christoph

-- 
Christoph Hormann <chr### [at] gmxde>
IsoWood include, radiosity tutorial, TransSkin and other 
things on: http://www.schunter.etc.tu-bs.de/~chris/

Post a reply to this message

> > hmm yes , but i m pretty sure bad situations will arise, for points near
the
> > edge of a cube for instance
> >
>
> As i understood the paper, these problems can be avoided by always using
> the same distance for calculating the diffusion term instead of using the
> individual distance of each sample.  It also says this is likely to
> produce some inaccuracy but it seems to work quite well.

i was not clear (sorry for my pathetic english, i wish i could draw a little
example :)
the problem is not for calculating the contribution of a sample but rather
to find the points which will contribute
how do you find points on the surface of the object , assuming the surface
is flat (but knowing this is not true) ?

M

Post a reply to this message

It seems to me that UV space could help here. For those unfamiliar, UV
mapping maps each point on the surface of an object to a point on a plane
with axes called u and v. In MegaPov, UV mapping is used to wrap a texture
around an object, and a UV mapper already exists for most primitives (not
pathological shapes like julia fractals) and for meshes. To find a point on
the surface near a given point, map the original point into UV space,
perturb the UV point a small amount, and then map back from UV onto the
object. As long as the UV map works, this point is guaranteed to be on the
surface of the object, no matter how convoluted the shape is.
"Mael" <mae### [at] hotmailcom> wrote in message
news:3bcd8e6d$1@news.povray.org...
> > > hmm yes , but i m pretty sure bad situations will arise, for points
near
> the
> > > edge of a cube for instance
> > >
> >
> > As i understood the paper, these problems can be avoided by always using
> > the same distance for calculating the diffusion term instead of using
the
> > individual distance of each sample.  It also says this is likely to
> > produce some inaccuracy but it seems to work quite well.
>
> i was not clear (sorry for my pathetic english, i wish i could draw a
little
> example :)
> the problem is not for calculating the contribution of a sample but rather
> to find the points which will contribute
> how do you find points on the surface of the object , assuming the surface
> is flat (but knowing this is not true) ?
>
> M
>
>

Post a reply to this message