POV-Ray : Newsgroups : povray.advanced-users : Potential speedup of isosurface rendering. Server Time
29 Mar 2024 10:47:38 EDT (-0400)
  Potential speedup of isosurface rendering. (Message 1 to 2 of 2)  
From: Michael Scharrer
Subject: Potential speedup of isosurface rendering.
Date: 21 Apr 2021 10:30:00
Message: <web.608035d6324774b68e75d824e4acb2c1@news.povray.org>
I just read the following paper: "Massively Parallel Rendering of
Complex Closed-Form Implicit Surfaces"
https://www.mattkeeter.com/research/mpr/

While it's not geared for ray tracing as is, I'm wondering if this method could
be used to add a pre-trace subdivision step as already exists for parametric
surfaces.

The most interesting aspect of the algorithm "tape shortening" which
automatically simplifies the function for the various segments. This seems to
work well for CSG heavy functions (min, max operations).

Or does Pov-Ray already do something along those lines?

--

Michael Scharrer
https://mscharrer.net/


Post a reply to this message

From: Le Forgeron
Subject: Re: Potential speedup of isosurface rendering.
Date: 21 Apr 2021 14:19:28
Message: <60806cb0$1@news.povray.org>
Le 21/04/2021 à 16:25, Michael Scharrer a écrit :
> I just read the following paper: "Massively Parallel Rendering of
> Complex Closed-Form Implicit Surfaces"
> https://www.mattkeeter.com/research/mpr/
> 
> While it's not geared for ray tracing as is, I'm wondering if this method could
> be used to add a pre-trace subdivision step as already exists for parametric
> surfaces.
> 
> The most interesting aspect of the algorithm "tape shortening" which
> automatically simplifies the function for the various segments. This seems to
> work well for CSG heavy functions (min, max operations).
> 
> Or does Pov-Ray already do something along those lines?
> 


The voxel split with tape shortening is nice, but seems a lot of
pre-work for a faster rendering later. Which is worth doing when
animating, but just moving the cost for a single image.

(and need an initial bounding of the surface: you cannot just split the
whole universe in only 2048³ voxels).

As with any GPU usage, the portability issues must also be raised.

IMHO, it is yet another move toward more memory usage to save some time.
The question being where is the right balance ?

Also, the theme being CSG, they go for the whole formula, and then use
voxels to reduce locally the complexity. Whereas the traditional CSG of
povray is to use bounding box to already ignore parts of the tree which
are irrelevant for a ray.

It is nice to have a huge formula with thousand operations, but it is
more readable to have a CSG in text form. Especially when you want to
get that hole a bit bigger, or rotate that other pilar around.


Post a reply to this message

Copyright 2003-2023 Persistence of Vision Raytracer Pty. Ltd.