In facestretch we describe how we exploited the dlib’s facial landmarks in order to measure face deformation and perform an expression recognition task. We implemented multiple approaches, based metric learning, neural networks and geodesic distances