Probably in 3.6.13 or 3.6.14 this problem will be fixed, then you don't need to disable Multiple Importance yourself.
@Fernando-Alcala Thanks for your comparisons, they are very helpful! Regarding the roughness matching issue you mentioned, Huang and Chiang use different roughness models. Huang's roughness…
Are you sure you disaled Multiple Importance on the light Area_90? When I disable it the result is 0.25, multiplied by 4/pi gives 0.318, which is expected.
Yes, it is because of this commit that I hit the assertion.
I hit an assertion
case DEVICE_CPU: |
|
/* This is the GPU denoiser - CPU devices shouldn't end up here. */ |
|
assert(0); |
This is due to CLOSURE_WEIGHT_CUTOFF
having a hard-coded value of 1e-5f
.
Closing as a duplicate of #117667.
Noticed when opening the file from blender/blender#123537 (comment)
For large scale scenes the quality is still not as good as in 4.1. I tried various approaches, for example computing a proper min_weight
, but they did not work well in smaller scenes.
In 4.2 we changed the sampling method in volume to only query the light tree once in volume segment (fdc2962bebb47e5d55564fcf30b4586f57f4311a). This does not work well when the volume has a large…