Discovering Discriminative Geometric Features with Self-Supervised Attention for Vehicle Re-Identification and Beyond



In the literature of vehicle re-identification (ReID), in-tensive manual labels such as landmarks, critical parts orsemantic segmentation masks are often required to improvethe performance. Such extra information helps to detect lo-cally geometric features as a part of representation learningfor vehicles. In contrast, in this paper, we aim to address the challenge ofautomaticallylearning to detect geometric fea-tures as landmarkswith no extra labels. To the best of ourknowledge, we are the first to successfully learn discrimi-native geometric features for vehicle ReID based on self-supervised attention. Specifically, we implement an end-to-end trainable deep network architecture consisting of threebranches:
(1) a global branch as backbone for image fea-ture extraction, (2) an attentional branch for producing at-tention masks, and (3) a self-supervised branch for regu-larizing the attention learning with rotated images to locategeometric features. We conduct comprehensive experimentson three benchmark datasets for vehicle ReID,i.e.,VeRi-776, CityFlow-ReID, and VehicleID, and demonstrate ourstate-of-the-art performance. We also show the good gener-alization of our approach in other ReID tasks such as per-son ReID and multi-target multi-camera (MTMC) vehicletracking.Our demo code is attached in the supplementaryfile.