[Bf-cycles] Volumetric patch update
kartochka22 at yandex.ru
Tue Jul 31 21:45:14 CEST 2012
Some progress with MCMC (MLT) sampler.
Ignoring B factor for now, average brightness must be tuned by hands,
using exposure. But it work, definitely, i found myself playing
re-rendering again and again all scenes with all that splashy mlt
looking when tiny dispersion used (< 0.0027).
MCMC works only for direct camera rays ( that started from light source,
and hit surface that have eval BSDF property, like diffussion BSDF), so
no glass or mirror, only diffusion mats and mayve gloss with roughness
(i dtill not sure it will work).
To turn that mode, use:
Max eye bounces:0
Max light bounces: 1.. 8 (clamped by 8 , need fix ugly quick hack later)
Mutations: >0, that is Markov chain serie length, try 10, or 1000 if
dare! (think it as samples number, more is slower, splashy, but smarter)
Startup: not implemented yet, ignored
Dispersion: 0.027 or less. Tricky thing, i use 0.0027 for mutations >
100, it looks more cool when converging, smoother splashes. 1.0 will
turn MCMC to something like dumb old sampler.
Only point light sources with radius 0 and emission meshes supported.
Of course, it work with volume, main test scene to debug MCMC is outdoor
"foggy street with 5-6 cone light"
Not for average tester usage, you must have some experience with MLT
theory, to make some image, better wait until i fix other path types.
The only reason i share that patch in almost broken state that it that
is first MLT picture i can make :), big deal for me.
To OL77, Brecht all other seriuos ppl: nothing useful yet, many
fireflies (i think color go negative then used in denominator), better
ignore, i will share cleaned version weeks later.
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 48610 bytes
Desc: not available
Url : http://lists.blender.org/pipermail/bf-cycles/attachments/20120731/5f32c8d9/attachment.bin
More information about the Bf-cycles