View-aware attribute guided network for vehicle re-identification

Abstract

Vehicle re-identification is one of the essential applications for intelligent transportation systems and urban surveillance. However, enormous variation in inter-class and intra-class resemblance creates a challenge for methods to distinguish between the same vehicles with different views. Additionally, diversified illumination and complicated environments create significant hurdles for the existing methods. We present a multi-guided learning method in this paper that uses multi-attribute and view point information, while also enhancing the robustness of feature extraction. The multi-attribute sub-network learns discriminative features like, i.e. color and type of vehicle. Moreover, the view predic-tor network adds extra information to the feature embedding and To validate the effectiveness of our framework, experiments on two benchmark datasets VeRi-776 and VehicleID are conducted. Experimental results illustrate our framework achieved comparative performance. VAAG CAR A CAR B Fig. 1 The figure shows two separate cars from distinct views from VeRi-776. Car A on the top row illustrates several views from various angles. Car B in the lower row has a diversity of viewpoints as well. These cars have very similar appearances, and it’s difficult to tell them apart. Car A appears to be different than its other views and identical to Car B from various perspectives.

Vehicle re-identification based on dimensional decoupling strategy and non-local relations

“…Recently, with the rise of neural networks and the proposal of large-scale datasets. Some methods use external resources [4][5][6][7] to capture discriminative features such as: part resolution, pose estimation, foreground segmentation, vehicle view labeling, etc. PVEN [4] uses a vehicle part parser to generate four different view masks, and then generates aligned local features by averaging pooling of the masks.…”