From Hand-Perspective Visual Information to Grasp Type Probabilities: Deep Learning via Ranking Labels