Weakly-Supervised Part-Attention and Mentored Networks for Vehicle Re-Identification

Vehicle re-identification (Re-ID) aims to retrieve images with the same vehicle ID across different cameras. Current part-level feature learning methods typically detect parts via uniform division, outside tools, or attention modeling. However, such part features often require expensive additional annotations and cause sub-optimal performance in case of unreliable mask predictions. In this paper, we propose a weakly-supervised Part-Attention Network (PANet) Part-Mentored (PMNet) for Re-ID. Firstly, PANet localizes part-relevant channel recalibration cluster-based generation without supervisory information. Secondly, PMNet leverages teacher-student guided distill part-specific from performs multi-scale global-part extraction. During inference, can adaptively extract discriminative localization by PANet, preventing unstable We address Re-ID issue as multi-task problem adopt Homoscedastic Uncertainty learn optimal weighing losses. Experiments are conducted on two public benchmarks, showing that our approach outperforms recent methods, which no extra an average increase 3.0% CMC@5 VehicleID over 1.4% mAP VeRi776. Moreover, method extend occluded task exhibits good generalization ability.