it's actually quite, like you expected.
for the eigen and fisherfaces fisherfaces model:
- num_components (to retain from the pca)
- the mean image
- the (pca-reduced set of) eigenvectors (multiplied by the lda eigenvec mat for fisherfaces)
- the eigenvalues (though they are never used in the algorithm)
- the projected train images
- the class labels
- the labels_info (name - id pairs)
for LBPH:LBPH:
- radius, neighbours, gridx, gridy (the lbp params, given in the constructor)
- the precomputed lbp-histograms from the train images
- the class labels
- the labels_info (name - id pairs)