it seems that you compute the centroids only once when initializing the model. But after training the model , the feature extracted from training data should have changed, the centroids should also change. Why not compute centroids in the training process?
it seems that you compute the centroids only once when initializing the model. But after training the model , the feature extracted from training data should have changed, the centroids should also change. Why not compute centroids in the training process?