Attentive Part-Based Alignment Network for Vehicle Re-Identification

Vehicle Re-identification (Re-ID) has become a research hotspot along with the rapid development of video surveillance. Attention mechanisms are utilized in vehicle Re-ID networks but often miss attention alignment across views. In this paper, we propose novel Attentive Part-based Alignment Network (APANet) to learn robust, diverse, and discriminative features for Re-ID. To be specific, order enhance discrimination part features, two part-level proposed APANet, consisting Part-level Orthogonality Loss (POL) (PAAL). Furthermore, POL aims maximize diversity via an orthogonal penalty among parts whilst PAAL learns view-invariant by means realizing fashion. Moreover, Multi-receptive-field (MA) module adopt efficient cost-effective pyramid structure. The structure is capable employing more fine-grained heterogeneous-scale spatial information through multi-receptive-field streams. addition, improved TriHard loss Inter-group Feature Centroid (IFCL) function optimize both inter-group intra-group distance. Extensive experiments demonstrate superiority our model over multiple existing state-of-the-art approaches on popular benchmarks.