How do I assess the quality of blob detection?

I have an image containing ellipses. I need to find precise centres for all the ellipses (order of 0.01 pixel).

I am using centroid finding and ellipse fitting algorithms to detect them and mark them. When I plot, I find that both of them are within .5 pixel but not sure which one is correct?

1) What method is best to get accuracies of order of 0.01 px ? 2) How do I say that my algorithm indeed achieved 0.01 px of accuracy?

please tell me statistical techniques to do so. I want to quantify.