Vehicle Re-Identification by Adversarial Bi-Directional LSTM Network

Abstract

Vehicle re-identification (re-ID) is an area that has received far less attention in the computer vision community than the prevalent person re-ID. Possible reasons for this slow progress are the lack of appropriate research data and the special 3D structure of a vehicle. Previous works have generally focused on limited views (e.g. front and rear), but these methods are less effective in realistic scenarios where vehicles usually appear in arbitrary views to cameras. In this paper, we focus on the uncertainty of vehicle viewpoint in re-ID, proposing an Adversarial Bi-directional LSTM Network (ABLN). Our model exploits the great advantages of the Long Short-Term Memory (LSTM) to model transformations across continuous view variations of a vehicle and adopts the adversarial architecture to enhance training. Thus, a global vehicle representation containing all views' information can be inferred from only one visible view, and then used for learning to measure the distance between two vehicles with arbitrary views. To verify our model, we evaluate the proposed method on the public VehicleID and VeRi datasets. Experimental results illustrate that our approach achieves consistent improvements over state-of-the-art vehicle re-ID methods.