Unsupervised Multi-Source Domain Adaptation Person Re-Identification Methods Based on Transformer and Graph Convolutional Neural Network

The Person Re-Identification task aims to utilize deep learning technology to determine whether specific pedestrians are in an image or video sequence. In the unsupervised Person Re-Identification task, how to extract effective feature representation is undoubtedly one of the main concerns. Although the existing methods have achieved excellent performances, the two perspectives about fine-grained detail information extraction and multi-source domain feature fusion remain the most significant challenges. Therefore, we introduce a new method of Person Re-Identification based on Transformer to extract and exploit the valuable data information. Firstly, we propose a baseline based on Transformer and use Transformer as the backbone network to extract richer feature information for the model training. This module can make the fine-grained features contain more spatial information to improve the reliability of the extracted instance features. Secondly, according to the multi-source domain, we design a domain information fusion module based on a graph convolutional neural network, which reduces the inter-domain differences by fusing the feature information of different domains extracted from the backbone network. Finally, extensive experiments on four benchmark datasets show that the proposed method outperforms most of the current unsupervised domain adaptation Person Re-Identification methods to a large extent.