A quick class I made to implement the eigenfaces algorithm with a reconstruction feature. Basically, after calculating the SVD of the database of faces, we can extract a basis for faces. This basis can then be used for a wide range of applications. My class allows you to give it any database of images and create this basis for its faces. Execute code_test.py if you want to have a quick test with the provided files. Don't hesitate to ask questions I'm always available !
On the training and testing sets respectively:
For this specific example, we took only 100 directions to recreate the eigenfaces, representing roughly 50 % of the variance of the faces.