Deep Metric Learning for Practical Person Re-Identification
This addresses the problem of practical person re-identification for surveillance and security applications, offering an incremental improvement over existing hand-crafted and metric learning methods.
The paper tackles person re-identification by proposing a deep metric learning method that learns similarity directly from image pixels using a siamese neural network, achieving superior results on VIPeR and PRID datasets in both intra-dataset and cross-dataset settings.
Various hand-crafted features and metric learning methods prevail in the field of person re-identification. Compared to these methods, this paper proposes a more general way that can learn a similarity metric from image pixels directly. By using a "siamese" deep neural network, the proposed method can jointly learn the color feature, texture feature and metric in a unified framework. The network has a symmetry structure with two sub-networks which are connected by Cosine function. To deal with the big variations of person images, binomial deviance is used to evaluate the cost between similarities and labels, which is proved to be robust to outliers. Compared to existing researches, a more practical setting is studied in the experiments that is training and test on different datasets (cross dataset person re-identification). Both in "intra dataset" and "cross dataset" settings, the superiorities of the proposed method are illustrated on VIPeR and PRID.