Deep Feature Fusion with Multiple Granularity for Vehicle Re-identification

Peixiang Huang, Runhui Huang, Jianjie Huang, Rushi Yangchen, Zongyao He, Xiying Li, Junzhou Chen

Abstract: Vehicle re-identification (Re-Id) plays a significant role in modern life. We found that Vehicle Re-Id and Person Re-Id are two very similar tasks in the field of Re-Id. To some extent, the Person Re-Id Networks can be transplanted to the Vehicle Re-Id tasks. In this paper, a Deep Feature Fusion with Multiple Granularity (DFFMG) method for Vehicle Re-Id is proposed for integrating discriminative information with various granularity. DFFMG is based on the Multiple Granularity Network (MGN), the state-of-the-art method from Person Re-Id. We pondered on the discrimination between Vehicle Re-Id and Person Re-Id. And we carefully designed DFFMG: a multi-branch deep network architecture which consists of one branch for global feature representations, two for vertical local feature representations and other two horizontal ones. Besides, several re-ranking methods were tested in our experiments and achieved higher scores. This network is adopted to train and test on the 2019 NVIDIA AI City Dataset [16]