Specific report and usage could be found in report, please read it for understanding the limitation of this project and the way to improve it.
Author: Yiling Liu
Student ID: 22214014
Original dataset for training(AT&T dataset with 40 faces)
Just for visualize the faces in train_faces
, it is a single file not necessary for data training, could be remove safely.
Runner to run .py
files in the current directory(except visualize_not_for_training.py
). Only used for combine those files together to run. In the real life schema, this packages will provide APIs for face recognition.
Generate an nary-tree for manage directories with .pgm
files inside it
Remove hair from the training picture to met the faces cropped by dlib
Sample data for training and put it into a .csv file. Notice this is just for project presentation, the better way is to store the numpy array in an .npy
file, so the computer do not need to load the .csv
file - this is an extra step!
Train a Siamese Network, generate model.h5
under dir models
Those are generated by .py
files mentions above
model.h5
is a trained model copied from build_model/models/model.h5
Runner
Use dlib to crop faces(If face in current picture is too small, skip it) from taken photo, then resize it to met the trained model. Use the existed model to compare the distance between faces in taken pictures.
Faces of valid people
Test inputs
Those are generated by .py
files mentions above