💡 This paper proposes SLRNet, a single-stage self-supervised low-rank network that learns robust low-rank pseudo-labels via cross-view collective matrix factorization, unifying weakly and semi-supervise