It is trivial for NEE. For forward tracing, if we rely on volume_stack_sample_method(kg, state)
then probably no additional extra path state is needed.
Although this does sound quite hacky and…
Attached is a scene where this PR has more noise with distance or MIS sampling. Not sure if this is realistic enough.
Ok I see your point, it is not directly related to equiangular/distance sampling, but light tree volume importance measure is biased towards equiangular sampling. Your suggestion would work when…
Seems to be fixed already. From the attached file I see nothing out of the ordinary, EEVEE and Cycles also match well.
The strength unit for mesh light is W/sr/m^2, integrated over the area and the hemisphere a factor of πA is expected, see blender/blender#108505 Can this…
Effectively we are still reusing the same random number to sample the light position from volume segment and from position, but that doesn't seem to cause problem as I observed. It was using the…
No. The change is that we determine the light to be sampled already in volume segment, not the light position. A new position would be sampled from the direct scattering position. Before this…