Fragment

[Frakkie]

Fragment



http://www.fractalforums.com/index.php?action=gallery;sa=view;id=13376


Fragmentarium, kalibox.

[cKleinhuis]

NICE one!! Global Illuminated?!?

[Jesse]

NICE one!! Global Illuminated?!?

Looks very nice and clean, so i assume it is not a full global illuminated image 

But i don't know, just messing around with this problem because to integrate it in a DE renderer seems not be very easy if you need a decent rendering time...
i attached an image with full relfects caustics made with a hdr ambient map and monte carlo method...
1200 rays per pixel in average and still very noisy due to the few light sources in the hdr background.
So this 2 primitives fast sample (non-fractal) approach took 15 mintes so far, 1 hour to halve this noise.

So i do a simplification for lightsources that has not all caustics but is way more faster, only the softed hardshadow areas must be computed more often...
but maybe Syntopia has better aproaches since i guess he his fiddling also with this feature 

[Syntopia]

Nice caustics!

So i do a simplification for lightsources that has not all caustics but is way more faster, only the softed hardshadow areas must be computed more often...
but maybe Syntopia has better aproaches since i guess he his fiddling also with this feature 

Not sure I've found any great revelations. You can do importance sampling (based on the cosine distribution for diffuse, and cosine-power distribution for specular), and stratified sampling, which lowers the noise and accelerates rendering, but not to the extent you are hoping for. I guess path tracing / monte carlo rendering just is slow, unless you use some funky spatial acceleration structures. 

I'm also having troubles with HDR-light maps with strong spot-lights. Combined with a rough fractal surface, it is almost impossible to avoid rendering artifacts.

[knighty]

Jesse is back! 
Excellent rendering. What kind of algorithm are you using?

I'm also having troubles with HDR-light maps with strong spot-lights. Combined with a rough fractal surface, it is almost impossible to avoid rendering artifacts.

Just my 2 cts: Maybe it's possible to take advantage of the Distance field to perform some kind of filtering.
 When one shoots a ray from a sample point to the environment map using sphere tracing, it's possible to obtain a cone with smallest aperture which doesn't intersect any "geometry" (an approximation to that cone could also do the job). The contributing light to that sample can be taken as the average of the light incoming from the environment map that is covered by the cone. Of couse, the average is weighted by the cosine terme but one can take it as a constant if the cone is not too wide. The computation of the avreage could be accelerated by using a mip map of the environment map.