[Bf-cycles] Volumetric patch update

storm kartochka22 at yandex.ru
Tue Aug 14 10:08:13 CEST 2012


Patched blender become even more slower, but looking at almost
firefly-free result, it promising.

The puzzle begins to take shape, still not ready for wide testing, but
much close then before. Fixed proper MIS between light ray and direct
light, many simple test scenes start looking firefly-free from 1 sample.
That take too much time, MIS logic really too complex for my programming
skills. Even with all papers, working examples (pbrt-family like
luxrender, mitsuba), i was expect it a bit simpler.

At last added naive volume material detection, needed for paths starting
from light (old logic can only handle paths starting from camera, and
camera must be in air, not inside any object). It work by firing
additional extra probe loop, as in shadow_blocked(), and calculate
current volume material from everything that was hit until background.
Obviously, it slowdown calculations, double it in worst case (low
max/min bounces number).

I played with MCMC sampling, learned a lot, now i understand better what
dispersion mean, it is spatial/angle precision of sampling details, so
in theory you must keep it as low as possible to not lose small details.
I suspect i need to rewrire MC mutation from angle based to spatial
based, to have equal steps (dispersion) in any 3d space, need more

I rethink logic behind min/max bounce number and "Max eye bonce"/"Max
light bounce". For now, keep min/max equal (to turn off russian roulete
decimation), assume it N, set "Max eye bonce" = N+2, set Max light
bounce" to N+1, all that need to be sure any bi-directional path sample
have corresponding alternate sampliing, to guarantee MIS work between
all paths. Later, i will remove it from GUI, and recalc internally by
current min/max.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: blender_cycles_volume_mlt_49883.patch.bz2
Type: application/x-bzip
Size: 54374 bytes
Desc: not available
Url : http://lists.blender.org/pipermail/bf-cycles/attachments/20120814/15966615/attachment.bin 

More information about the Bf-cycles mailing list