ProFeat: Unsupervised image clustering via progressive feature refinement

Jeonghoon Kim, Sunghoon Im, Sunghyun Cho

Abstract: Highlights•We propose a novel iterative approach to unsupervised image clustering.•We rigorously divide representation learning and clustering.•We progressively refine representations using confident samples.•We can learn discriminative features avoiding influence from incorrect clusters.•We also propose an ensemble-based approach.