FaceLib: Face Analysis

Used for face detection, facial expression, AgeGender estimation and recognition with PyTorch.

• Instalation: pip install git+https://github.com/sajjjadayobi/FaceLib.git

How to use:

Check this example_notebook or take a look at the following sections

1. Face Detection: RetinaFace

You can use these backbone networks: Resnet50, mobilenet. Default model is mobilenet and it will be automatically downloaded.

• The following example illustrates the ease of use of this package on your webcam:

fromfacelibimportWebcamFaceDetectordetector=WebcamFaceDetector()
detector.run()

• Low-level access to bounding boxes and face landmarks

fromfacelibimportFaceDetectordetector=FaceDetector()
boxes, scores, landmarks=detector.detect_faces(image)

2. Face Alignment: Using face landmarkd

For face aligment always use the detect_align function it gives you better performance.

• Face detection and aligment using the detect_align function.

fromfacelibimportFaceDetectordetector=FaceDetector()
faces, boxes, scores, landmarks=detector.detect_align(image)

Original	Aligned & Resized	Original	Aligned & Resized

3. Age & Gender Estimation:

ShufflenetFull is the default model, and it will be automatically downloaded.

• Age and gender estimation live on your webcam (or any camera)

fromfacelibimportWebcamAgeGenderEstimatorestimator=WebcamAgeGenderEstimator()
estimator.run()

• Low-lvel access to ages and genders

fromfacelibimportFaceDetector, AgeGenderEstimatorface_detector=FaceDetector()
age_gender_detector=AgeGenderEstimator()
faces, boxes, scores, landmarks=face_detector.detect_align(image)
genders, ages=age_gender_detector.detect(faces)
print(genders, ages)

4. Facial Expression Recognition:

The default model is densnet121 and it will be automatically downloaded. Note that face size must be (224, 224).

• Emotion detector live on your webcam

fromfacelibimportWebcamEmotionDetectordetector=WebcamEmotionDetector()
detector.run()

• Emotions as an array with their probabilities

fromfacelibimportFaceDetector, EmotionDetectorface_detector=FaceDetector(face_size=(224, 224))
emotion_detector=EmotionDetector()
faces, boxes, scores, landmarks=face_detector.detect_align(image)
emotions, probab=emotion_detector.detect_emotion(faces)

• on my Webcam 🙂

5. Face Recognition: InsightFace

• This module is a pytorch reimplementation of Arcface(paper), or Insightface(Github)

Pretrained Models & Performance

• IR-SE50

LFW(%)	CFP-FF(%)	CFP-FP(%)	AgeDB-30(%)	calfw(%)	cplfw(%)	vgg2_fp(%)
0.9952	0.9962	0.9504	0.9622	0.9557	0.9107	0.9386

• Mobilefacenet

LFW(%)	CFP-FF(%)	CFP-FP(%)	AgeDB-30(%)	calfw(%)	cplfw(%)	vgg2_fp(%)
0.9918	0.9891	0.8986	0.9347	0.9402	0.866	0.9100

Prepare the Facebank

Save the images of the faces you want to detect in this folder

Insightface/models/data/facebank/
---> person_1/
---> img_1.jpg
---> img_2.jpg
---> person_2/
---> img_1.jpg
---> img_2.jpg

You can save a new preson in facebank with 2 ways:

• Use add_from_webcam: it takes 4 images and saves them on facebank.

fromfacelibimportadd_from_webcamadd_from_webcam(person_name='sajjad')

• use add_from_folder: it takes a path with some images from just a person.

fromfacelibimportadd_from_folderadd_from_folder(folder_path='./', person_name='sajjad')

Recognizer

The default model is mobilenet and it will be automatically downloaded

• Face Recognition live on your webcam

fromfacelibimportWebcamVerifyverifier=WebcamVerify(update=True)
verifier.run()

• Low-level access to your images

importcv2fromfacelibimportFaceRecognizer, FaceDetectorfromfacelibimportupdate_facebank, load_facebank, special_draw, get_configconf=get_config()
# conf.use_mobilenet=False # if you want to use the bigger modeldetector=FaceDetector(device=conf.device)
face_rec=FaceRecognizer(conf)
# set True when you add someone new to the facebankupdate_facebank_for_add_new_person=Falseifupdate_facebank_for_add_new_person:
targets, names=update_facebank(conf, face_rec.model, detector)
else:
targets, names=load_facebank(conf)
image=cv2.imread(your_path)
faces, boxes, scores, landmarks=detector.detect_align(image)
results, score=face_rec.infer(faces, targets)
print(names[results.cpu()])
foridx, bboxinenumerate(boxes):
special_draw(image, bbox, landmarks[idx], names[results[idx]+1], score[idx])

Reference: InsightFace