Iterative 3D shape classification by online metric learning

To provide a scalable and flexible tool for 3D shape classification, this paper proposes an iterative 3D shape classification method by integrating incrementally updating, online learning and user intervention. It classifies the collection of 3D shapes iteratively by combing unsupervised clustering with online metric learning, and puts the user intervention into the loop of each iteration. The features of our method lie in three aspects. Firstly, it discovers the potential groups in the collection group by group without any pre-labeled samples or any pre-trained classifiers. Secondly, the users can get the desirable classes by directly confirming the required members of each group and annotate them with any free label to suit for different applications. Finally, the scalable collection can be handled dynamically and efficiently by our incrementally updating mechanism based on the online metric learning. The experimental results prove the effectiveness of the proposed method.