I'm using python and the Olivetti dataset (with a lot of 64x64 grayscale face images).
I'm creating a matrix 20x4096, where each line is a sample (I chose 20 samples to be the training set) and each column is a value between 0 and 255, representing all the 64x64 pixels.
After doing PCA in this matrix, I get something about 10 eigenvectors (with 4096 slots each).
The eigenvectors are all normalized (meaning their length is 1). How can an eigenvector represent an eigenface? All its values are something around 0.001 or so.
