YOLOv7 Face

Introduction

YOLOv7 Face is a Python library to perform face detection and anonymization using YOLOv7 models. If employed with a good model, it can facilitate implementing face detection and anonymization for surveillance cameras and other computer vision applications.

Install

The easiest way to install the YOLOv7Face is by using pip:

pip install git+https://github.com/msamsami/yolov7face

Getting Started

Take the following steps to set your face detector and anonymizer up and running:

Define model configurations

First, you need to define your face detector model configurations by initializing and instance of YOLOv7Configs class. This class requires a weights parameter which can be set in multiple ways. If you already have a pre-trained YOLOv7 model, simply pass the path to its *.pt file:

from yolov7face import YOLOv7Configs

model_configs = YOLOv7Configs(weights='my_models/face_detector.pt', cfg='widerface.yaml')

If the *.pt file is available online rather than your local drive, or you want to keep your model information (e.g., name, version, URL, author, etc.) in a maintainable way, you may want to use the YOLOv7Model class which is used to maintain YOLOv7 models. When both filepath and url are passed to this class, if the model object is not found in the filepath, it will be automatically downloaded from the URL and saved to the filepath. After defining your model as a YOLOv7Model instance, you can pass it to the weights parameter of YOLOv7Configs:

from yolov7face import YOLOv7Model, YOLOv7Configs

my_model = YOLOv7Model(
    name="Awesome Face Detector",
    filepath='my_models/awesome_face_detector.pt',
    url="https://mystorage.blob.core.windows.net/models/yolov7-face-detector.pt",
    version="0.0.14",
    author="Johnny Cash"
)

model_configs = YOLOv7Configs(weights=my_model, cfg='widerface.yaml')

In case you don't have your own pre-trained face detector model, you can use one of the pre-trained models that comes with the library:

from yolov7face import YOLOV7_WIDERFACE_MODEL, YOLOv7Configs

model_configs = YOLOv7Configs(weights=YOLOV7_WIDERFACE_MODEL, cfg='widerface.yaml')

Define anonymizer configurations

NOTE: If you don't want to apply any anonymization to detected faces, you can skip this step.

In this step, you should define the face anonymizer configuration if you want to anonymize the faces detected by the model. This can be done by creating an instance of the FaceAnonymizer class. The face anonymizer currently supports three types of anonymization:

• Blurring (method='blur')
• Pixelation (method='pixelate')
• Block-out (method='block')

from yolov7face import FaceAnonymizer

anonymizer = FaceAnonymizer(method='blur', blur_kernel_size=(19, 19), blur_sigma_x=45, blur_sigma_y=45)

Inference

Face detection

Now that face detector configurations are defined, you can initialize your face detector by creating an instance of the YOLOv7Face class and using it for inference:

from yolov7face import YOLOv7Face, YOLOv7Configs, YOLOV7_WIDERFACE_MODEL

model_configs = YOLOv7Configs(weights=YOLOV7_WIDERFACE_MODEL)
detector = YOLOv7Face(configs=model_configs)

Inference on an image:

detector.predict_img('meeting.jpg')

Inference on a video:

detector.predict_video('classroom.mp4', save_to='classroom-faces.mp4')

Live inference on webcam:

detector.predict_webcam()

Face anonymization

To anonymize detected faces, you should pass the face anonymizer instance to the anonymizer parameter of YOLOv7Face:

from yolov7face import YOLOv7Face, YOLOv7Configs, YOLOV7_WIDERFACE_MODEL, FaceAnonymizer

model_configs = YOLOv7Configs(weights=YOLOV7_WIDERFACE_MODEL)
anonymizer = FaceAnonymizer(method='blur', blur_kernel_size=(19, 19), blur_sigma_x=45, blur_sigma_y=45)
detector = YOLOv7Face(configs=model_configs, anonymizer=anonymizer)

Anonymize faces in an image:

detector.predict_img('meeting.jpg')