This paper proposes a strong convolutional baseline (PCB) and refined part pooling (RPP) for person retrieval. PCB employs a uniform partition strategy to learn part-informed features, achieving competitive results with state-of-the-art methods. RPP refines the uniform partition by reassigning outliers to the parts they are closest to, enhancing within-part consistency. On the Market-1501 dataset, PCB achieves 92.3% rank-1 accuracy and 77.4% mAP, while RPP further improves these metrics to 93.8% and 81.6%, respectively. PCB is a simple and effective baseline, and RPP significantly enhances its performance by improving within-part consistency. The paper also compares PCB with other methods, including variants of the part classifier and attention mechanisms, and shows that PCB with RPP outperforms them. The results demonstrate that PCB and RPP are effective for person retrieval, achieving state-of-the-art performance on multiple datasets.This paper proposes a strong convolutional baseline (PCB) and refined part pooling (RPP) for person retrieval. PCB employs a uniform partition strategy to learn part-informed features, achieving competitive results with state-of-the-art methods. RPP refines the uniform partition by reassigning outliers to the parts they are closest to, enhancing within-part consistency. On the Market-1501 dataset, PCB achieves 92.3% rank-1 accuracy and 77.4% mAP, while RPP further improves these metrics to 93.8% and 81.6%, respectively. PCB is a simple and effective baseline, and RPP significantly enhances its performance by improving within-part consistency. The paper also compares PCB with other methods, including variants of the part classifier and attention mechanisms, and shows that PCB with RPP outperforms them. The results demonstrate that PCB and RPP are effective for person retrieval, achieving state-of-the-art performance on multiple datasets.