Disentangling the underlying feature attributes within an image with no prior\nsupervision is a challenging task. Models that can disentangle attributes well\nprovide greater interpretability and control. In this paper, we propose a\nself-supervised framework DisCont to disentangle multiple attributes by\nexploiting the structural inductive biases within images. Motivated by the\nrecent surge in contrastive learning paradigms, our model bridges the gap\nbetween self-supervised contrastive learning algorithms and unsupervised\ndisentanglement. We evaluate the efficacy of our approach, both qualitatively\nand quantitatively, on four benchmark datasets.\n