I have a dataset which has images with variability in face posture, lighting, and occlusions. I want to recognize facial expressions in real time using the camera. I want to train a model using FER dataset. But due to variation in face postures and occlusions, the classifier will not be able to classify the expressions with good accuracy. I want to detect the face parts like eyes, nose, and mouth which can help to recognize the expressions. Please help me and guide me that how to solve this problem?? How to extract face parts from only frontal images of FER dataset???