Graph Cut-guided Maximal Coding Rate Reduction for Learning Image Embedding and Clustering

Wei He, Zhiyuan Huang, Xianghan Meng, Xianbiao Qi, Rong Xiao, Chun-Guang Li; Proceedings of the Asian Conference on Computer Vision (ACCV), 2024, pp. 1883-1899

In the era of pretrained models, image clustering task is usually addressed by two relevant stages: a) to produce features from pretrained vision models; and b) to find clusters from the pre-traiend features. However, these two stages are often considered separately or learned by different paradigms, leading to suboptimal clustering performance. In this paper, we propose a unified framework for jointly learning structured embeddings and clustering, termed graph Cut-guided Maximal Coding Rate Reduction (CgMCR^2), in which the learning of clustering results effectively facilitates the learning of embeddings toward forming a union-of-orthogonal-subspaces. To be specific, in CgMCR^2, we integrate a flexible and principled clustering module into the framework of maximal coding rate reduction, in which the clustering module provides partition information to help the cluster-wise compression for the embeddings and the learned embeddings in turn help to yield more accurate clustering results. We conduct extensive experiments on both standard and out-of-domain image datasets and experimental results validate the effectiveness of our approach.