<p>Person Re-Identification (Re-ID) aims to retrieve specific pedestrian images from various cameras. Most Re-ID studies have concentrated on pedestrian images captured by visible cameras, often neglecting infrared images taken in low-light conditions. However, Visible-Infrared Person Re-Identification (VI-ReID) is crucial for enhancing public safety. In addition, existing methods for extracting pedestrian features using a vision Transformer are relatively simple. In order to extract more discriminative pedestrian representations, we propose a multi-granular Vision Transformer (MGVT) for (VI-ReID). Specifically, we design a multi-granular feature extraction baseline for Visible-Infrared Person Re-Identification, which contains three main stages: Stage 1 (replacement), Stage 2 (fusion), and Stage 3 (feature learning), which in turn effectively helps the model to improve its accuracy. In addition, we also design a dynamic enhancement block, which can dynamically enhance the discriminative features of pedestrians so that the network can focus on more useful features and filter out some redundant features. Compared with the latest state-of-the-art methods, our method yields 2.05% and 5.17% increase in R1 identification on SYSU-MM01 dataset and RegDB dataset, respectively.</p>