[Training]:
        [Step-4]:
            Feeding the pos(1),neg(0) & ignore(-1) images of size 12 x 12 to PNet.
            (Note: landmark samples are not trained at this stage)
        [Step-5]:
            PNet predicts objectness(whether a face is present or not) & bbox of face.
            Loss function is optimized in Pnet & finally output PNet model.