You must predict the location of various facial features in images. We recommend reshaping the input data into an image shape and then running convolutional filters. The last layer should be a regression layer.