VisageSnap

English | 한국어

Recognizes faces and trains models, brings in the pictures and provides identification predictions and face classification. It also performs semi-supervised learning.

Feature

• Recognize faces.
• Train the model through semi-supervised learning with labeled or unlabeled pictures.
• Predicts if the face belongs to someone it knows and whose face it is.

Installation

Requirements

• Python 3.9+
  • Versions below 3.9 have not been tested, and pickle module must be installed via pip.
• dilb

First, you need to install dilb. You can install it by following the instructions on the here.

Then, you can install VisageSnap by using pip:

pip install visagesnap

Usage

Assign a label to the face

You should assign a NameLabel to the face you want to classify first.

vs = VisageSnap.Core()
people = ['Tom', 'Jerry']
# ['NameLabel1', 'NameLabel2', 'NameLabel3'...]

vs.set_label(people)

You can also do it like this so that assign a NameLabel and NumberLabel: (NumberLabel MUST NOT BE -1)

people = {
    # 'NameLabel': NumberLabel
    'Tom': 0,
    'Jerry': 1
}

Put the picture files to be used during training in the directory. In this case, the file name follows the following rules:

(NameLabel)-(Any character).extension

Tom-123.png
Tom-124.jpg
Tom-126.jpeg
Jerry-2.png
Jerry-3.png
Jerry-4.png

Recognize faces and train the model

Before training, you need to load model. You don't have a model? It's okay. It'll be created automatically if it doesn't exist.

vs.load_model()

Train with the picture files in the directory.

vs.train_labeled_data()

If you want to train with unlabeled data, you can also try to like this:

vs.train_unlabeled_data()

Identification predictions

Put the picture files you want to predict into the directory.

result = vs.predict_all()
print(result)

{
   "target1.png": "Tom",
   "target2.jpeg": "Jerry",
   "target3.jpeg": ["Tom", "Jerry"], # multiple faces in one picture
   "target4.jpeg": None # If the face is unknown
}

Full example

You can see the full code here.

To change the directory you work with

You should put the picture files into configured directory, and also model file is stored in model directory.

from VisageSnap import Directory

my_dir = Directory( "labeled_pic",
                    "unlabeled_pic",
                    "my_model.d",
                    "predict_dir")
vs.set_directory(my_dir)

Default Directory:

{
    "labeled": "labeled",
    "unlabeled": "unlabeled",
    "model": "model"
}

Others

To get the face information

You can get the faceObject by using get_faceObject method. The faceObject is a dataclass that contains the face information.

from visagesnap import From
# From.LABEL, From.FILENAME

face_tom = vs.get_faceObject(From.LABEL, "Tom")
face_tom = vs.get_faceObject(From.FILENAME, "Tom-123.png")
# face_tom = Face(label, encodings, filenames)

name: str = face_tom.label
encodings: NDArray = face_tom.encodings
filenames: list = face_tom.filenames

To change prediction threshold

vs.threshold = 0.5  # Default: 0.48

Acknowledgement

• scikit-learn
• face_recognition