Vehicle Re-Identification with Refined Part Model

Vehicle re-identification (re-ID) is a challenging task in computer vision. Different vehicle identities belonging to the same model may have a similar appearance with little inter-instance discrepancy, while one vehicle may have large intra-instance differences under different viewpoint and illumination. In this paper, we propose a refined part model to learn an efficient feature embedding to solve this problem. Different from other methods, which directly obtain region part for vehicle re-ID. The refined part model is formulated through a Grid Spatial Transformer Network (GSTN) to automatically locate the vehicle and perform division for regional features. Residual attention is also conducted to give an additional refinement for a fine-grained identification. Finally, the refined part features are fused to form an efficient feature embedding. Extensive experimental results show that our approach significantly outperforms state-of-the-art on two large scale datasets: VehicleID and Vehicle-1M.