first we use the data of faces & angles (yaw,pitch,roll) AFLW2000 ,and we use mediapipe to extract 468 point of each face then combine the two into one data frame which consists of the points and the angles
second we train three models (randomforest) to estimate the angles , used random search grid to help finding best models
finally we draw the axis to see the results on images,video and webcam