Generalized Zero Shot Learning

In the field of computer vision, Generalized Zero-Shot Learning (GZSL) refers to dividing the set of categories into seen and unseen classes. The training process utilizes the visual representations of the seen classes and the semantic features of both seen and unseen classes. During the testing phase, the model uses the visual representations of both seen and unseen classes to achieve recognition and classification of unseen categories. GZSL aims to address the issue of insufficient data annotation and enhance the model's generalization ability, making it highly valuable for practical applications.