Jointly Learn the Base Clustering and Ensemble for Deep Image Clustering
Chen Liang, Zhiqian Dong, Sheng Yang, Peng Zhou · 2024
Deep image clustering attracts increasingly more attention in computer vision and multimedia communities. To tackle the stableness and robustness problems in clustering, clustering ensemble is applied to generate a better result by fusing multiple weak base clustering results. However, the existing deep clustering ensemble methods only focus on how to ensemble multiple fixed weak base results and ignore the influence of the consensus result on the base results. Alternatively, in this paper, we propose another question, i.e., how to use the ensemble to improve the base results? To this end, we present a novel joint deep clustering ensemble framework, which jointly generates the base results and does the ensemble, so that the deep clustering and the clustering ensemble can boost each other. In this framework, we design a base clustering generation module and an ensemble module and integrate them into a unified neural network architecture. The extensive experiments on benchmark datasets well demonstrate the effectiveness and superiority of the proposed method. The code is available at https://github.com/liangchen98/JDCE.