Nighttime Person Re-Identification via Collaborative Enhancement Network with Multi-domain Learning

Andong Lu,Tianrui Zha,Chenglong Li,Jin Tang,Xiaofeng Wang,Bin Luo

DOI: https://doi.org/10.48550/arXiv.2312.16246

2023-12-25

Computer Vision and Pattern Recognition

Abstract:Prevalent nighttime ReID methods typically combine relighting networks and ReID networks in a sequential manner, which not only restricts the ReID performance by the quality of relighting images, but also neglects the effective collaborative modeling between image relighting and person ReID tasks. To handle these problems, we propose a novel Collaborative Enhancement Network called CENet, which performs the multilevel feature interactions in a parallel framework, for nighttime person ReID. In particular, CENet is a parallel Transformer network, in which the designed parallel structure can avoid the impact of the quality of relighting images on ReID performance. To perform effective collaborative modeling between image relighting and person ReID tasks, we integrate the multilevel feature interactions in CENet. Specifically, we share the Transformer encoder to build the low-level feature interaction, and then perform the feature distillation to transfer the high-level features from image relighting to ReID. In addition, the sizes of existing real-world nighttime person ReID datasets are small, and large-scale synthetic ones exhibit substantial domain gaps with real-world data. To leverage both small-scale real-world and large-scale synthetic training data, we develop a multi-domain learning algorithm, which alternately utilizes both kinds of data to reduce the inter-domain difference in the training of CENet. Extensive experiments on two real nighttime datasets, \textit{Night600} and \textit{RGBNT201$_{rgb}$}, and a synthetic nighttime ReID dataset are conducted to validate the effectiveness of CENet. We will release the code and synthetic dataset.