This paper proposes a novel data-driven method called Saliency-Guided Patch Transfer (SPT) for occluded person re-identification (ReID). SPT leverages real-world scenes in the training set to achieve controllable occlusion construction. The method divides the sample after patchify into an identity set and an occlusion set through salient patch selection. By recombining these two subsets, SPT can effectively exploit scene information from the dataset and produce high-quality occluded samples. Furthermore, an occlusion-aware Intersection over Union (IoU) with mask rolling and a class-ignoring training strategy are proposed to control SPT's process, ensuring stable and effective patch transfer. Consequently, SPT can be seamlessly integrated into ViT-based algorithms, resulting in significant performance improvements in occluded ReID. SPT is designed to address the issue of occlusion in person ReID, where traditional methods assume the entire body of a person is visible. However, in real-world applications, occlusions caused by obstacles or other persons can significantly degrade performance. SPT uses a vision transformer to divide person instances and background obstacles using salient patch selection. By transferring person instances to different background obstacles, SPT can easily generate photo-realistic occluded samples. Additionally, an occlusion-aware Intersection over Union (OIoU) with mask rolling is introduced to filter the most suitable combination and a class-ignoring strategy is used to achieve more stable processing. Extensive experimental evaluations on occluded and holistic person ReID benchmarks demonstrate that SPT provides a significant performance gain among different ViT-based ReID algorithms on occluded ReID. The method is evaluated on several datasets, including Occluded-Duke, Occluded-REID, Market-1501, and DukeMTMC-reID. The results show that SPT achieves competitive performance on both occluded and holistic ReID datasets. The method is also compared with previous state-of-the-art methods, and it is shown that SPT can significantly improve the performance of ViT-based ReID algorithms in occluded scenarios. The proposed method is effective in generating realistic occluded samples and improving the robustness of ReID models against occlusions.This paper proposes a novel data-driven method called Saliency-Guided Patch Transfer (SPT) for occluded person re-identification (ReID). SPT leverages real-world scenes in the training set to achieve controllable occlusion construction. The method divides the sample after patchify into an identity set and an occlusion set through salient patch selection. By recombining these two subsets, SPT can effectively exploit scene information from the dataset and produce high-quality occluded samples. Furthermore, an occlusion-aware Intersection over Union (IoU) with mask rolling and a class-ignoring training strategy are proposed to control SPT's process, ensuring stable and effective patch transfer. Consequently, SPT can be seamlessly integrated into ViT-based algorithms, resulting in significant performance improvements in occluded ReID. SPT is designed to address the issue of occlusion in person ReID, where traditional methods assume the entire body of a person is visible. However, in real-world applications, occlusions caused by obstacles or other persons can significantly degrade performance. SPT uses a vision transformer to divide person instances and background obstacles using salient patch selection. By transferring person instances to different background obstacles, SPT can easily generate photo-realistic occluded samples. Additionally, an occlusion-aware Intersection over Union (OIoU) with mask rolling is introduced to filter the most suitable combination and a class-ignoring strategy is used to achieve more stable processing. Extensive experimental evaluations on occluded and holistic person ReID benchmarks demonstrate that SPT provides a significant performance gain among different ViT-based ReID algorithms on occluded ReID. The method is evaluated on several datasets, including Occluded-Duke, Occluded-REID, Market-1501, and DukeMTMC-reID. The results show that SPT achieves competitive performance on both occluded and holistic ReID datasets. The method is also compared with previous state-of-the-art methods, and it is shown that SPT can significantly improve the performance of ViT-based ReID algorithms in occluded scenarios. The proposed method is effective in generating realistic occluded samples and improving the robustness of ReID models against occlusions.