paper and code issue
thanks for your work.I have a confusion: in your paper, you describe performing normalization of the covariance matrix before clustering, but in the code, it seems that clustering is done before normalization.Hope your reply.
The renormalization is applied after the cluster step.
Where in the paper does it say that the normalization is performed before the clustering?
“Clustering is then performed using the normalized covariance matrices, and each Gaussian stores, in addition to a codebook index, the scalar scaling factor η.”
Here normalized refers to the gaussians having a scale vector of length one (scaling factor η is extracted from each matrix). After every clustering step they are re-normalized to make sure the matrices stay normalized.