Lets say I render 200 images with different seed values and 50 samples each. I wrote a program that adds all color values of each pixel and color together and then divides it by the amount of images. So am I allowed to say, that the combined images has 10.000 samples?
$\begingroup$
$\endgroup$
7
-
$\begingroup$ Yes, almost. This technique isn't quite as noiseless as rendering the equivalent number of samples at once. $\endgroup$– gandalf3Commented Sep 12, 2015 at 6:57
-
$\begingroup$ So is there any formula, for how many samples it would be as a straight render? $\endgroup$– rudolf97Commented Sep 12, 2015 at 7:01
-
$\begingroup$ See blender.stackexchange.com/a/2808/599. If I understand correctly, I think this would depend on the scene and how well samples are distributed. The stacking technique is more or less equivalent to straight Monte Carlo integration. The reason cycles does slightly better is because it uses Quasi-Monte Carlo integration, which tries to avoid duplicate samples. The increase in efficiency with Quasi-Monte Carlo integration is apparently difficult to calculate.. $\endgroup$– gandalf3Commented Sep 12, 2015 at 7:18
-
$\begingroup$ Ok... That seems quite complicated... Because I do not wanna say 10000 samples, if it equals for example only 5000 samples... $\endgroup$– rudolf97Commented Sep 12, 2015 at 8:22
-
$\begingroup$ You can say that you approximate 10,000 samples by combining multiple images of less samples. $\endgroup$– samblerCommented Sep 12, 2015 at 13:33
|
Show 2 more comments