An End-to-end Clustering and Classification Learning Network

Abstract

With the dramatic increase of image data, image clustering algorithms have received extensive attention. Meanwhile, deep learning has shown excellent performance in various fields, while there are few studies on its application in image clustering. Moreover, previous deep clustering frameworks are multistage and difficult to embed the available supervision information. In view of the above, we propose a deep clustering network called CCLN (Clustering and Classification Learning Network), which combines feature learning, clustering, and classification to achieve end-to-end learning. At the same time, we add a branch of the supervisory to utilize multi-Grained supervision information, so as to improve the performance of image clustering. Extensive experiments show that CCLN effectively utilizes available supervision information, and improves clustering performance.