Proposing CiPR framework for better representation learning in GCD through cross-instance positive relations.