NightIR‐ReID: An Infrared Nighttime Person Re‐Identification Benchmark

ABSTRACT

Person re‐identification (Re‐ID) plays a vital role in intelligent surveillance systems by enabling the recognition of individuals across multiple camera views. Although substantial progress has been made in daytime Re‐ID, achieving robust performance under nighttime conditions remains challenging due to the low‐light environment and severe image noise. To further explore the challenges of nighttime Re‐ID, we present NightIR‐ReID , a large‐scale infrared Re‐ID dataset collected using mobile surveillance devices in real‐world nighttime scenarios. The dataset contains 785 identities and 35,166 images captured under diverse and complex conditions, including various adverse weather scenarios, providing a comprehensive benchmark for studying nighttime Re‐ID challenges. Furthermore, we propose the compensation and denoising framework that aims to enhance the robustness of Re‐ID representations under challenging nighttime conditions. This framework integrates a denoising branch and an illumination compensation branch to extract complementary and mutually reinforcing feature representations. Experimental results validate the effectiveness of our approach and highlight the significance of NightIR‐ReID in advancing research in this field. NightIR‐ReID will be released at https://github.com/Qingming‐Leng/NightIR‐ReID .