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

File Structure

dir build_model

dir train_faces

Original dataset for training(AT&T dataset with 40 faces)

visualize_not_for_training.py

Just for visualize the faces in train_faces, it is a single file not necessary for data training, could be remove safely.

__main__.py

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.

filetree.py

Generate an nary-tree for manage directories with .pgm files inside it

cut_train.py

Remove hair from the training picture to met the faces cropped by dlib

file_processor.py

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!

build_model.py

Train a Siamese Network, generate model.h5 under dir models

other files and directories

Those are generated by .py files mentions above

dir detect_face

dir models

model.h5 is a trained model copied from build_model/models/model.h5

__main__.py

Runner

compare.py

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.

dir valid_faces

Faces of valid people

dir test

Test inputs

other files and directories

Those are generated by .py files mentions above