scott <sco### [at] laptopcom> wrote:
> Write it in OpenGL or DirectX.  A modern machine should be able to draw 
> 10000 spheres in real-time (ie at least 30fps), with per-pixel specular and 
> diffuse lighting.

  Are you sure about that?

  How many triangles do you need to make a triangle mesh to look even
remotely as a sphere? Multiply that by 10000.

-- 
                                                          - Warp

Post a reply to this message

"Rahul"  wrote in message
<web.47dd807ac23a6d959eba08c0@news.povray.org>:
> My scene is basically nothing more than a large number of spheres(of different
> sizes and colors) Povray works great for me for rendering an excellent high
> quality final output.
> 
> Unfortunately there are parts in my development cycle where I need a faster
> render. Low quality acceptable. Are there any options people would recommend?

Do your sphere intersect?

If they do not, handling the hidden surfaces is pretty easy: the nearest is
in front of the farthest.

One possible solution would be to generate a SVG file, with radial gradients
to simulate lightning and specular light, sorting the spheres upon the
distance to the camera. Then let Inkscape or anything similar rasterize it.

> I dont even need the full ammunition of a raytracer. But something that could
> act on  a similar-to-povray file would be ideal. (or something that a povray
> ...pov file could easily be parsed into)

You said your POV file was itself generated: I would recommend to use the
data that was used to generate it instead of re-parsing it.

Post a reply to this message

"Rarius" <rar### [at] rariuscouk> wrote:

> Might I suggest you post an example POV file to povray.binaries.scene-files
> so we can all have a play and see if we can speed it up.

Thanks Rarius! I'll post a copy sometime later today. Meanwhile I'll also try to
use your suggestions. Its not that many spheres really. I strongly suspect I've
a coding inefficiency somewhere. Probably a blunder; I'm sort of a povray
newbie.

In fact, the render time is "LARGE" only relatively. Its a few seconds really
but that becomes slow when (a) I start thinking in terms of designing a GUI (b)
when I have a string of molecular images to render in an interactive setup.

BTW, I'm pleasantly surprised to see that many other posters actually thought
raytracing might be the *fastest* option around!

-Rahul

Post a reply to this message

"scott" <sco### [at] laptopcom> wrote:

> Write it in OpenGL or DirectX.  A modern machine should be able to draw
> 10000 spheres in real-time (ie at least 30fps), with per-pixel specular and
> diffuse lighting.


That's what I was thinking of. Do I have to start from scratch(C/ Java)? Or are
there any meta-scripting languages. I don't think I really need all the OpenGL
functionality here.

Here's the more general question though: Povray at its basic level is based on a
scene definition language and many combinations of CSG primitives. These seem to
be somewhat fundamental operations with no particular tie to the "raytracing"
method. Or am I wrong here?

Are there other codes out there that can operate on a "scene definition
language" and render a scene albeit with a "shortcut" visualization method.

-Rahul

Post a reply to this message

> Are there other codes out there that can operate on a "scene definition
> language" and render a scene albeit with a "shortcut" visualization method.
> 

You should see if this (old) POV-Ray OpenGL Patch does what you need, at 
an acceptable speed:

http://www.daylongraphics.com/other/povray/patches/

It supports wireframe for a bunch of objects, and solid rendering only 
for spheres (with *no* textures, spheres always show in gray).

http://www.daylongraphics.com/other/povray/patches/#solid

Post a reply to this message

Nicolas George <nicolas$george@salle-s.org> wrote:
> Do your sphere intersect?

No. They do not. None intersect.

> One possible solution would be to generate a SVG file, with radial gradients
> to simulate lightning and specular light, sorting the spheres upon the
> distance to the camera. Then let Inkscape or anything similar rasterize it.

I like the SVG idea. That's the sort of thing I was thinking about. The SVG
renderer can handle the details! I only want to give it my scene definition.
I'll look up the SVG docs now; before I've only coded 2D stuff in SVG.


> You said your POV file was itself generated: I would recommend to use the
> data that was used to generate it instead of re-parsing it.

True. I should do that. I have access to the raw data. Which is nothing but
tuples of (x,y,z, radius, color)

Sorry, its probably much easier if I post the raw pov file here. But its
intermediate output. So I need to play a bit to clean it up and isolate it.

-Rahul

Post a reply to this message

> I like the SVG idea. That's the sort of thing I was thinking about. The SVG
> renderer can handle the details! I only want to give it my scene definition.
> I'll look up the SVG docs now; before I've only coded 2D stuff in SVG.
> 

SVG only supports 2D. But if you do your own 3D->2D projection, 
rendering that 2D image would probably be faster than raytracing.

Post a reply to this message

Rahul wrote:
> BTW, I'm pleasantly surprised to see that many other posters actually thought
> raytracing might be the *fastest* option around!

Scanline rendering starts with a blank screen, then draws shapes one 
polygon at a time. Its speed tends to be limited by the number of shapes 
it has to draw, so the many spheres scene is basically the worst-case 
for scanline.

Raytracing draws a screen one pixel at a time, regardless of how many 
shapes there are. Its speed is more complicated, but is usually limited 
by a combination of resolution (number of pixels it has to draw) and the 
number of recursive calls it has to make (shadows and reflections).

BTW, random thought--have you tried making all the light sources shadowless?

-- 
William Tracy
afi### [at] gmailcom -- wtr### [at] calpolyedu

    You can never go back. Free Mars.
     -- Kim Stanley Robinson, _Green Mars_

Post a reply to this message

Warp wrote:
>   How many triangles do you need to make a triangle mesh to look even
> remotely as a sphere? Multiply that by 10000.

Warp is basically right, but come to think of it, a cube (or 
tetrahedron) might be a suitable approximation for a sphere in this 
case. :-)

-- 
William Tracy
afi### [at] gmailcom -- wtr### [at] calpolyedu

    "Meanwhile we can't be having a constitution that says only 'don't 
change speed too fast.'"
     -- Kim Stanley Robinson, _Blue Mars_

Post a reply to this message

> BTW, random thought--have you tried making all the light sources 
> shadowless?
> 

He tried +q2 and got no noticeable speedup. q2 disables shadows.

Post a reply to this message