Cross-Domain Person Re-Identification Using Heterogeneous Convolutional Network

Person re-identification (Re-ID) is a challenging task due to variations in pedestrian images, especially cross-domain scenarios. The existing person Re-ID approaches extract the feature from single image, but they ignore correlations among images. In this paper, we propose Heterogeneous Convolutional Network (HCN) for Re-ID, which learns appearance information of images and simultaneously. To end, first utilize Neural (CNN) features Then construct graph target dataset where are treated as nodes similarity represents linkage between nodes. Afterwards, Dual Graph Convolution (DGConv) explicitly learn correlation similar dissimilar samples, could avoid over-smoothing caused by fully connected graph. Furthermore, design HCN multi-branch structure mine structural pedestrians. We conduct extensive evaluations on three datasets, i.e. Market-1501, DukeMTMC-reID MSMT17, results demonstrate that superior state-of-the-art methods.