Generalized Zero-Shot Learning

In Generalized Zero-Shot Learning (GZSL), the category set is divided 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, while the testing phase evaluates the visual representations of both seen and unseen classes simultaneously. GZSL aims to improve the model's recognition ability for unseen categories, extending the application scope of traditional machine learning methods, and thus has significant research and application value.