LAPNet: Local Aware Permutation NetWork for Vehicle Re-Identification

Vehicle re-identification is a pivotal element in contemporary intelligent transportation systems. Nonetheless, existing vehicle re-identification methodologies frequently encounter a substantial challenge: the limited information encapsulated in query images, which often depict only a partial aspect of the vehicle. This significantly exacerbates the complexity of matching vehicle images captured from varied perspectives. To tackle this issue, this paper introduces an avant-garde Local Aware Permutation Network. This proposed network adeptly perceives discriminative local regions in vehicle images and meticulously fine-tunes features essential for the identification process. Moreover, it integrates an advanced information exchange mechanism, facilitating the transfer and sharing of data between different sub-local features, thus markedly enhancing the network’s capability to extract and represent local features. Through exhaustive comparative experiments and ablation studies on the VeRi-776 and VehicleID datasets, we have rigorously validated the efficacy and superiority of the proposed method.