Reuse the sampled light position would make it easier for equiangular sampling to compute the valid range, not sure if it would be noisier or cause fireflies near the light source, I can also give…
I'm not sure I like this solution, but it's the first one I got working without bias (except for the aforementioned TODOs). A potential problem is that the result might be noisier when the light…
You should remove the computation of sin_b_sin_c_2 now that it's unused. Otherwise looks good.
Turns out we are using the same random number for volume segment and volume direct light. Without light tree it means we sample the same light twice, with light tree the two samples are correlated,…
Thank you! This is much appreciated.
"Blend factor between an octave and its previous one. A value of zero corresponds to zero detail"
This is rather hard to understand. Maybe "The difference in scale between two consecutive octaves. A larger value corresponds to a larger difference".
..give more detailed noise but increase render time.