Be Precise or Fuzzy: Learning the Meaning of Cardinals and Quantifiers from Vision