Will a poorly chosen random algorithm degrade the quality of the rendered picture?
This might be answered in the Monte Carlo lecture, but why did we choose to take the absolute value of the dot product of wi and the normal, instead of some other method like allowing the negative or clipping at 0?
It seems like this is very sequential, are there ways to parallelize this?
If we have to make recursive calls with this sampling method, then how is it an improvement from computing the exact integral of the rendering equation? Is computing the integral that much more computationally intensive?
Why don't we assume that each sample has the same probability?