Closed Easy-Shu closed 3 years ago
Each FLAME shape and expression components are scaled by the standard deviation. Therefore, varying the shape and expression parameters in the range [-n, n] corresponds to a variation of [-nsigma, nsigma]. In order to get the eigenvalues back, reshape the shapedirs (5023 x 3 x 400) to a matrix of shape 15069 x 400, and get the norm of 15069-dimensional columns of the matrix. This norm corresponds to the standard deviation, which is the square root of the eigenvalue of the covariance matrix. Does this answer your question?
Thanks for your reply. I think I can calculate the eigenvalue of the covariance matrix.
The FLAME2020 model only provides the principle components (PCs), but where is the corresponding eigenvalues?