Abstract
Vehicle Re-Identification (Re-ID), which aims at matching vehicle identities across different cameras, is a critical technique for traffic analysis in a smart city. It suffers from varying image quality and challenging visual appearance characteristics. A solution for enhancing the feature robustness is by training Convolutional Neural Networks on multiple datasets simultaneously. However, the larger set of training data does not guarantee performance improvement due to misaligned feature distribution between domains. To mitigate the domain gap, we propose a Joint Domain Re-Identification Network (JDRN) to improve the feature by disentangling domain-invariant information and encourage a shared feature space between domains. With our JDRN, we perform favorably against state-of-the-arts methods on the public VeRi-776 dataset and obtain promising results on the 2019 AI City Challenge.