Re: Update on Diffusion Curves

On 2011-09-28 22:38, Sebastian Markbåge wrote:
> On Mon, Mar 7, 2011 at 10:35 AM, Jasper van de Gronde 
> <th.v.d.gronde@hccnet.nl <mailto:th.v.d.gronde@hccnet.nl>> wrote:
> ...
>
> I did a WebGL implementation of Jeschke's model last year.
>
> It did make me concerned about standardizing Diffusion Curves (such as 
> in SVG).
>
> Some artifact may not be visible in one implementation while it is 
> clearly visible in another. When rendering images designed using the 
> original tool by Orzan et al. I get some artifacts where the artist 
> have left imperfections. Those imperfections are not visible in 
> Orzan's implementation. Likewise, Jeschke's model hides other 
> artifacts that ARE visible in Orzan's model.
>
> http://labs.calyptus.eu/diffusioncurves/
>
> What makes Diffusion Curves unique is that a small artifact can have a 
> huge impact on the over all image, as it's spread out by the solver.
If you're talking about diffusion curves in the idealized sense, you are 
right that in principle all boundary points influence the entire image, 
but in practice the influence usually tapers off quite quickly. However, 
most solvers do have a bit of a problem with actually converging to a 
solution, which led me to my thesis subject. One problem is that all 
solvers tend to give smooth solutions that sort of look plausible, so it 
is not immediately obvious that something is wrong. In my thesis I 
describe a potential solution, by using a fundamentally different kind 
of solver, but it's definitely not a drop-in replacement for existing 
solvers (at least not yet).

Another way of attacking the problem is to redefine diffusion curves to 
behave differently. Whether there is indeed a "better" definition is 
very much an open question.

(I couldn't view your site btw, for some reason I can't get webgl to 
work, possibly because my notebook hardly has anything worthy of the 
name GPU.)