We employ a learning-based method for the 3D localization of fetal face, i.e. finding a bounding box for the fetal
face. The learning-based method takes box detection as a two-category classification problem: each candidate box either contains the target object or not.