Deeply-Learned Spatial Alignment for Person Re-Identification

Dongyue Chen,Mengjiao Cao,Xiaosheng Yu,Peng Chen,Tong Jia

doi:10.1109/access.2019.2945353

Dongyue Chen, Mengjiao Cao + Show 3 more

Open Access

https://doi.org/10.1109/access.2019.2945353

Copy DOI

Abstract

A large class of Person Re-identification (ReID) approaches identify pedestrians with the TriHard loss. Though the TriHard loss is a robust ReID method, pose variance and viewpoint in pedestrians constrain the performance. To address this problem, we introduce a spatial transformer network (STN) to align pedestrians. Then, we illustrate the generality of the STN module in pose variance problem through the evaluations on feature representation network (FRN) like VGG, ResNet and DenseNet architectures respectively. Furthermore, based on the evaluation results, we propose a robust and high-performance ReID model which consists of the STN module, DenseNet backbone and TriHard loss. And finally, we prove that our ReID model is whole differentiable by formula derivation, therefore achieving an end-to-end high-performance ReID system. The experiments show that our ReID system outperforms the state-of-art methods on Market-1501, DukeMTMC-reID and CUHK03 datasets.

Highlights

Person Re-identification (ReID) is a branch of image retrieval issue which aims to identify pedestrian of interest across non-overlapping gallery images
We propose a robust and high-performance ReID model which consists of the spatial transformer network (STN) module, DenseNet backbone and the TriHard loss
Based on the empirical results, we propose a robust and high-performance ReID model which consists of the STN module, DenseNet backbone and TriHard loss

Summary

Introduction

Person Re-identification (ReID) is a branch of image retrieval issue which aims to identify pedestrian of interest across non-overlapping gallery images. It is an important research because of the great value in social life like security monitoring, traffic management and crime detection. Most of the deep learning ReID approaches use ResNet50 [1] as the backbone and fine-tune on the ImageNet [2] pre-trained model. Metric learning method has turned into a loss function in the deep ReID model. The TriHard approach trains CNN to get feature representations and computes them by TriHard loss, performing an end-to-end metric learning method for ReID issue

Objectives

Findings

Methods