First of all thanks for providing the nice implementation of the Marching Squares algorithm.

I've noticed some inconsistencies for various edge cases (literal "edge"-cases). I'm testing a simple 4x4 grayscale input data where all the values are 0 except one of the corners and a threshold of 0.5. Depending on which corner is set to 1, the number of mesh points varies between 3, 5, and 8. As a result, the boundary extraction also gives significantly different results of what (I think) should just be shaped which are rotated versions of each other.