Retraining the human pose estimation model

We will now discuss how to handle videos and retrain our human pose estimation network. We have already covered face detection and how to apply a model to a video. Opening a video is pretty straightforward and OpenCV provides a mechanism for that. It's basically doing the same thing one frame at a time. The following example shows the code for this:

predictor_path = "./shape_predictor_68_face_landmarks.dat"
detector = dlib.get_fronta1_face_detector()
predictor = dlib.shape_predictor(predictor_path)

#Uncomment Line below if you want to use your webcam
#cap = cv2.VideoCapture(0) #0 is the first camera on your computer, change if you
#more than one camera

#Comment out the Line below if using webcam
cap = cv2.VideoCapture('./rollerc.mp4')
figure(100)
font = cv2.FONT_HERSHEY_SIMPLEX

First, we need to create a cv2 capture device...