Transductive Zero-Shot Learning with Visual Structure Constraint

However, it is unrealistic to label all the object classes, thus making these supervised learning methods struggle to recognize objects which are unseen during training.