We are planning to use facial landmark information as input to the model. Since there are more than 60 points, it doesn\'t look good to use 60 channels as inputs after one-h
