While supervised techniques in re-identification are extremely effective, the\nneed for large amounts of annotations makes them impractical for large camera\nnetworks. One-shot re-identification, which uses a singular labeled tracklet\nfor each identity along with a pool of unlabeled tracklets, is a potential\ncandidate towards reducing this labeling effort. Current one-shot\nre-identification methods function by modeling the inter-relationships amongst\nthe labeled and the unlabeled data, but fail to fully exploit such\nrelationships that exist within the pool of unlabeled data itself. In this\npaper, we propose a new framework named Temporal Consistency Progressive\nLearning, which uses temporal coherence as a novel self-supervised auxiliary\ntask in the one-shot learning paradigm to capture such relationships amongst\nthe unlabeled tracklets. Optimizing two new losses, which enforce consistency\non a local and global scale, our framework can learn learn richer and more\ndiscriminative representations. Extensive experiments on two challenging video\nre-identification datasets - MARS and DukeMTMC-VideoReID - demonstrate that our\nproposed method is able to estimate the true labels of the unlabeled data more\naccurately by up to $8\\%$, and obtain significantly better re-identification\nperformance compared to the existing state-of-the-art techniques.\n