It's interesting how choosing a sampling distribution that does not exactly match f(x) would still sort of work since it would correct itself by amplifying sampled values that were unlikely to be sampled.
s3kim2018
I think a sampling distribution p2 will increase noise in our sample, but it will make it so that if N is large enough, the lighting effects of the scene will be dimmed (as we will be sampling more from darker rays). Maybe this is how photos are dimmed instead of ambiently dimming the whole picture? I don't think this is the most efficient way to dim a photo though.
It's interesting how choosing a sampling distribution that does not exactly match f(x) would still sort of work since it would correct itself by amplifying sampled values that were unlikely to be sampled.
I think a sampling distribution p2 will increase noise in our sample, but it will make it so that if N is large enough, the lighting effects of the scene will be dimmed (as we will be sampling more from darker rays). Maybe this is how photos are dimmed instead of ambiently dimming the whole picture? I don't think this is the most efficient way to dim a photo though.