Multi-task Mid-level Feature Alignment Network for Unsupervised Cross-Dataset Person Re-Identification

Most existing person re-identification (Re-ID) approaches follow a supervised framework, in which a large number of labelled matching pairs are for training. Such a setting severely limits their scalability in-world applications where no labelled samples are available during the phase. To overcome this limitation, we develop a novel unsupervised-task Mid-level Feature Alignment (MMFA) network for the unsupervised-dataset person re-identification task. Under the assumption that the and target datasets share the same set of mid-level semantic attributes, proposed model can be jointly optimised under the person's identity and the attribute learning task with a cross-dataset mid-level alignment regularisation term. In this way, the learned feature can be better generalised from one dataset to another which improve the person re-identification accuracy. Experimental results on benchmark datasets demonstrate that our proposed method outperforms the-of-the-art baselines.