Researcher(s): Sangyeon Kim, Yoojin An

Zero-shot learning (ZSL) is to predict a class not seen in the training phase. ZSL can be achieved in many ways. The most effective ZSL method is an attribute-based one. In the attribute-based ZSL, each class is represented by a prototype, a set of attributes. Therefore, a model has to correctly extract the prototype from an image to predict its class for the zero-shot classification.