README.md

Face identification using CNN + TPE demo

Face identification demo which implements the Triplet Probabilistic Embedding for Face Verification and Clustering work.

Requirements

• python3
• keras
• dlib
• scikit-image

Usage

Get all the files from here and put them to the model dir. Then start demo_app1.py.

Training

Download the face_template.npy and shape_predictor_68_face_landmarks.dat from here and put them to the model dir.

Place training data in following order:

data\
    dev_protocol.npy
    dev\
        1.jpg
        2.jpg
        3.jpg
        ...
    test\
        subject_0\
            1.jpg
            2.jpg
            ...
        subject_1\
            1.jpg
            2.jpg
            ...
        ...
    train\
        subject_0\
            1.jpg
            2.jpg
            ...
        subject_1\
            1.jpg
            2.jpg
            ...
        ...

Then run as follows:

1. utils/load_data.py
2. train_cnn.py
3. train_tpe.py

Use the test scripts to test your model.