Cell instance segmentation in Pap smear image remains challenging due to the\nwide existence of occlusion among translucent cytoplasm in cell clumps.\nConventional methods heavily rely on accurate nuclei detection results and are\neasily disturbed by miscellaneous objects. In this paper, we propose a novel\nInstance Relation Network (IRNet) for robust overlapping cell segmentation by\nexploring instance relation interaction. Specifically, we propose the Instance\nRelation Module to construct the cell association matrix for transferring\ninformation among individual cell-instance features. With the collaboration of\ndifferent instances, the augmented features gain benefits from contextual\ninformation and improve semantic consistency. Meanwhile, we proposed a sparsity\nconstrained Duplicate Removal Module to eliminate the misalignment between\nclassification and localization accuracy for candidates selection. The largest\ncervical Pap smear (CPS) dataset with more than 8000 cell annotations in Pap\nsmear image was constructed for comprehensive evaluation. Our method\noutperforms other methods by a large margin, demonstrating the effectiveness of\nexploring instance relation.\n