Christoph Hormann wrote:
> 
> Mark Weyer wrote:
> 
>>I do not see why the quality is better than using repulsion
>>inside the unit circle with a uniform force field before projecting.
> 
> 
> To me it seems fairly obvious that projecting a disc on a hemisphere does
> not maintain the distance relations and therefore reduces the uniformity
> of a distribution.  

The point is that you do not want a uniform distribution on the 
hemisphere ...

> 
> To be precise, the points at the rim of the hemisphere will be quite dense
> in direction of the circumference but will be moved apart in theta
> direction.

Ahh! If you used a unit circle to do the repulsion thing then you are 
quite right - you do not get an even distribution at the edge: all the 
outermost points would be pressed against the rim.

This is why I used a 2 radius circle when I was doing the Delaunay 
triangulation, and adding samples until I had enough in the unit circle. 
That way you get an even distribution right across the unit circle - 
right to the rim of the hemisphere.

You could do the same with a repulsion algorithm - fill a circle of 
greater than 1 radius and keep adding a sample and repelling to 
completion untill you have the right number of samples within the 1 unit 
circle.

> 
> Christoph
> 

Mike Andrews.

Post a reply to this message

Michael Andrews wrote:
> 
> [...]
> 
> Ahh! If you used a unit circle to do the repulsion thing then you are
> quite right - you do not get an even distribution at the edge: all the
> outermost points would be pressed against the rim.
> 
> This is why I used a 2 radius circle when I was doing the Delaunay
> triangulation, and adding samples until I had enough in the unit circle.
> That way you get an even distribution right across the unit circle -
> right to the rim of the hemisphere.

The radius of the circle does not matter, once you project it on a
hemisphere you get an anisotropic distribution (meaning the density of the
points depends on the direction like explained in my last post).  But feel
free to try it out, i'm sure if you check the quality of the results you
will see it.

Christoph

-- 
POV-Ray tutorials, include files, Sim-POV,
HCR-Edit and more: http://www.tu-bs.de/~y0013390/
Last updated 15 Nov. 2002 _____./\/^>_*_<^\/\.______

Post a reply to this message

Christoph Hormann wrote:
> The radius of the circle does not matter, once you project it on a
> hemisphere you get an anisotropic distribution (meaning the density of the
> points depends on the direction like explained in my last post).  But feel
> free to try it out, i'm sure if you check the quality of the results you
> will see it.

This anisotropy is inevitable - the density of the samples is decreasing 
  with theta but constant in phi. If you plot the radiosity sample 
positions you can see this quite clearly with the official samples.

> 
> Christoph
> 

Mike Andrews.

Post a reply to this message

Rohan Bernett wrote:
>I tried these settings and the scene slowed down terribly. As I'm writing
>this, the render has been going 56min and is only on line 258.

Which sounds reasonably fast... esp. if you consider that these ones could
serve as not too bad settings for the final render.

>The lo-res
>preview and what's done so far looks pretty good, although there are some
>odd blotches near the light sources.
>
>I hate having to wait ages for just a test render. I think anything over 5
>minutes for an 800*600 image with no AA, is too slow.

hmm I've waited hours for 320x240 test renders (and we all know of users who
wait days for that)... what's the matter? When I build up a scene, I use a
simpler lighting setup, no radiosity (or very low settings), no reflections
etc. As soon as the final render begins, it could last days, I couldn't
care less (as long as the scene is worth it...)

>I hate slow renders.

Everybody does, but remember, first comes quality.

--
Jonathan

Post a reply to this message

> Ahh! If you used a unit circle to do the repulsion thing then you are
> quite right - you do not get an even distribution at the edge: all the
> outermost points would be pressed against the rim.
> 
> This is why I used a 2 radius circle when I was doing the Delaunay
> triangulation, and adding samples until I had enough in the unit circle.
> That way you get an even distribution right across the unit circle -
> right to the rim of the hemisphere.

I am uneasy about this. I fear you do not solve the problem
but only make it less visible by moving the edge farther out.
I would prefer doing the thing without any edges at all: Use
the unit square as a torus, find an even distribution there,
map it onto the unit circle via
  phi <- 2*pi*x
  r <- sqrt(y)
and then proceed as you suggested.

  Mark

Post a reply to this message

Michael Andrews wrote:
> I suppose it could be, and I've thought about doing this. However you 
> would have to calculate an array for each level of radiosity recursion 
> because the code reduces the number of samples taken at each subsequent 
> recursion level. And the results wouldn't be very different anyway.

I've had another think about this, and I don't think it would work in 
some cases.

The code uses the apparent (texture modified) surface normal and checks 
whether a particular sample can be used or is going behind the actual 
surface. If the sample can not be used then it takes a fresh sample.

So you can not know a priori how many samples you need at a particular 
recursion level because it may alter at each sampling point.

This is why you not only need an even sample distribution but also to 
have the distribution as even as possible over the first N samples of 
the distribution where N can be any number up to the maximum possible 
number of samples.

And this is why I was looking at an iterative Delaunay triangulation - 
to get a 'good enough' even distribution for the first N samples of the 
distribution.

For something like a repulsion algorithm which sets the position of all 
the sample points at once you would then need to choose the sampling 
order carefully to get a good distribution for any number of points.

Enough babbling ...

Bye for now,
	Mike Andrews.

Post a reply to this message

Michael Andrews wrote:
> 
> [...]
> 
> This is why you not only need an even sample distribution but also to
> have the distribution as even as possible over the first N samples of
> the distribution where N can be any number up to the maximum possible
> number of samples.

Exactly, like i said in

news://news.povray.org/3DE76813.8B1CB162%40gmx.de

> For something like a repulsion algorithm which sets the position of all
> the sample points at once you would then need to choose the sampling
> order carefully to get a good distribution for any number of points.

Yes, that's the idea how to do it.  Generate a number of point using an
arbitrary algorithm (but chosen for a uniform but non-regular
distribution) and then sort them for maximum uniformity at lower sample
counts too.  I can't think think of a different method that does not
produce a regular distribution.

Christoph

-- 
POV-Ray tutorials, include files, Sim-POV,
HCR-Edit and more: http://www.tu-bs.de/~y0013390/
Last updated 15 Nov. 2002 _____./\/^>_*_<^\/\.______

Post a reply to this message

>hmm I've waited hours for 320x240 test renders (and we all know of users who
>wait days for that)

Man is that slow!

>When I build up a scene, I use a
>simpler lighting setup, no radiosity (or very low settings), no reflections
>etc.

I don't think I can simplify my lighting setup. I only have the neccessary
number of light sources to match the locations in the real world scene. As
for no radiosity, for the flouro lights that's a no-can-do.

>As soon as the final render begins, it could last days, I couldn't
>care less (as long as the scene is worth it...)

Only if you have a second computer to handle the slow rendering.

As for test renders, hours or days is far too long. For test renders, 10
minutes is getting too slow, I'd definatly prefer 30 second test renders.

I should probably stop grumbling and go and read a different thread now.

Rohan _e_ii

Post a reply to this message

On Wed,  4 Dec 2002 03:51:07 EST, "Rohan Bernett" <rox### [at] yahoocom>
wrote:

>As for test renders, hours or days is far too long. For test renders, 10
>minutes is getting too slow, I'd definatly prefer 30 second test renders.

I think you should either stick to simple scenes or use another
program :) But even in "fast" renderers like MAX, a scene with
raytraced soft shadows, reflections, radiosity and so on will not fall
within your expectations of a render time.


Peter Popov ICQ : 15002700
Personal e-mail : pet### [at] vipbg
TAG      e-mail : pet### [at] tagpovrayorg

Post a reply to this message

>I think you should either stick to simple scenes or use another
>program :) But even in "fast" renderers like MAX, a scene with
>raytraced soft shadows, reflections, radiosity and so on will not fall
>within your expectations of a render time.

Lucky for me I'm using soft shadows, reflections and radiosity, yet.
Although by the time I get to the final render, I probably will be. That'll
probably mean a render of several days, given the ambitious complexity of
the intended final scene.

Rohan _e_ii

Post a reply to this message