Closed dkit closed 9 years ago
In the paper they use a perturbation as big as the march increment, [-march_increment/2 ... march_increment]. That implementation yields the following results, where the image on the top is without jittering and the one in the bottom is with jittering. Would you like to have a controllable knot for the noise or is it enough with a constant noise scale?
The previous image was generated jittering the origin randomly, I noticed that I misread the technique and it had to be done along the ray direction. However that approach doesn't modify the output image in any noticeable way.
Check out section 3.2 in this paper: https://people.irisa.fr/Ricardo.Marques/articles/2009/GPURayCasting-EPCG2009.pdf
I used it in some of my experiments (with my own ray caster) and it works pretty well, although it sometimes took some fiddling to generate the right noise level...
The reason I bring this up is if I view the flame from the side and get too close then you can clearly see the voxels. We could probably just use higher resolutions, but maybe this would work too....