Different Covariance Structures for Naive Bayes Classifier

Question

I am trying to build a naive bayes classifier to classify handwritten digits (mnist).

I am currently following similar logic to this tutorial where the variance they