[Bf-committers] Blender and QMC sampling

Magnus Löfgren lofgrenmeister at gmail.com
Sun Jun 13 11:10:47 CEST 2010


Thanks for clearing that up Matt :) I guess the complete de-gutting of the
poor renderpipeline won't happen for a looong time, good thing someone's
working on a 3delight exporter then ;) :o ;)

But there is one thing that I suspect isn't an enormous task, wich will make
the Full Sample Motion blur option ALOT more optimized.

Would an option to change the sampling of area shadows, glossy reflections,
AO etc to random sampling in between the motion samples/sub-frames work?
Currently the noise patterns are identical in between sub-frames/motion
samples/passes, so it doesn't take advantage of the fact that the whole
image is re-rendered X amount of times.

If the noise were more random, you could basically decrease the amount of
shadow/AO/reflection samples by the same X amount of times you re-render the
image, and the combined end-result would be virtually noise-free given
enough motion blur passes :)

Project messiah uses this (http://www.projectmessiah.com), I think they just
divide the raytrace samples automatically by the amount of subframes (but
user control would be preferable), and it works really well!


Afterall blender has 3d motion blur, it's just not very optimized :)
One could probably find even more ways to optimize it in addition to the
sampling stuff.
Whachathink? :)


More information about the Bf-committers mailing list