Class-Incremental Novel Class Discovery

We study the new task of class-incremental Novel Class Discovery (class-iNCD), which refers to problem discovering novel categories in an unlabelled data set by leveraging a pre-trained model that has been trained on labelled containing disjoint yet related categories. Apart from classes, we also aim at preserving ability recognize previously seen base Inspired rehearsal-based incremental learning methods, this paper propose approach for class-iNCD prevents forgetting past information about classes jointly exploiting class feature prototypes and feature-level knowledge distillation. self-training clustering strategy simultaneously clusters trains joint classifier both classes. This makes our method able operate setting. Our experiments, conducted three common benchmarks, demonstrate significantly outperforms state-of-the-art approaches. Code is available https://github.com/OatmealLiu/class-iNCD .