since these are the same, is there a relationship between the size of the convolution kernel you're using in Option #1, and the number of super-samples you're taking to find the average in Option #2 ?