I understand MC integration is favorable in the cases where the integrand is high dimensional, and the more samples we draw the less noise we can expect from our results since the variance of error will decrease.
Now my question is that is there any rule when should we stop increasing the number of samples to do MC? Concretely, if we write a MC shading program, and has #samples as a hyperparameter to tune to get a nice image, is trial-and-error the only way of determining this parameter?
Will we be covering rejection sampling? This is mentioned in more depth in the review session but we don't appear to go over it in lecture.