SparCC correlation with different sample sizes


I have a question about SparCC. If there are two correlation networks with different sample sizes (N=65 and N=21) to be compared, do I need to use Fisher's z-transformation to normalize for sample sizes? Or SparCC already did? Because like Pearson's correlation coefficient, Fisher's z-transformation would be applied to correct.
(formula = 1/2ln((1+r)/(1-r))(1/(N-3)^0.5, r=SparCC value, N=sample size)