Face Recognition

Simple library to recognize faces from given images

Face Recognition pipeline

Below the pipeline for face recognition:

• Face Detection: the MTCNN algorithm is used to do face detection
• Face Alignement Align face by eyes line
• Face Encoding Extract encoding from face using FaceNet
• Face Classification Classify face via eculidean distrances between face encodings

How to install

pip install git+https://github.com/paoloripamonti/face-recognition

How to train custom model

Initialize model

from face_recognition import FaceRecognition

fr = FaceRecognition()

Train model with pandas DataFrame:

fr = FaceRecognition()

fr.fit_from_dataframe(df)

where 'df' is pandas DataFrame with column person (person name) and column path (image path)

Train model with folder:

fr = FaceRecognition()

fr.fit('/path/root/')

the root folder must have the following structure:

root\
    Person_1\
        image.jpg
        ...
        image.jpg
    Person_2\
        image.jpg
        ...
        image.jpg
    ...
        

Save and load model

you can save and load model as pickle file.

fr.save('model.pkl')

fr = FaceRecognition()

fr.load('model.pkl')

Predict image

fr.predict('/path/image.jpg')

Recognize faces from given image. The output is a JSON with folling structure:

{
  "frame": "image data", # base64 image with bounding boxes
  "elapsed_time": time, # elapsed time in seconds
  "predictions": [
      {
        "person": "Person", # person name
        "confidence": float, # prediction confidence
        "box": (x1, y1, x2, y2) # face bounding box
      }
  ]
}

Example

For more details you can see: https://www.kaggle.com/paoloripamonti/face-recogniton