This paper addresses the challenge of person re-identification (Re-ID) by proposing an end-to-end feature learning strategy that integrates discriminative information at multiple granularities. The authors introduce the Multiple Granularity Network (MGN), a multi-branch deep network architecture that consists of one branch for global feature representations and two branches for local feature representations. Unlike previous part-based methods that focus on specific pre-defined semantic regions, MGN uniformly partitions images into several stripes and varies the number of parts in different local branches to obtain local feature representations with multiple granularities. This approach allows the network to capture both coarse and fine-grained discriminative information, enhancing the robustness and accuracy of Re-ID systems. The MGN architecture is designed to learn global and local features independently, with each branch dedicated to a specific level of granularity. The global branch captures integral but coarse features, while the local branches focus on more detailed and fine-grained representations. The network employs softmax loss for classification and triplet loss for metric learning, ensuring robust and discriminative feature representations. Extensive experiments on mainstream datasets such as Market-1501, DukeMTMC-reID, and CUHK03 demonstrate that MGN achieves state-of-the-art performance, outperforming existing methods by a significant margin. The method's effectiveness is further validated through ablation studies, which show that the multi-branch architecture and the use of triplet loss are crucial for achieving superior results.This paper addresses the challenge of person re-identification (Re-ID) by proposing an end-to-end feature learning strategy that integrates discriminative information at multiple granularities. The authors introduce the Multiple Granularity Network (MGN), a multi-branch deep network architecture that consists of one branch for global feature representations and two branches for local feature representations. Unlike previous part-based methods that focus on specific pre-defined semantic regions, MGN uniformly partitions images into several stripes and varies the number of parts in different local branches to obtain local feature representations with multiple granularities. This approach allows the network to capture both coarse and fine-grained discriminative information, enhancing the robustness and accuracy of Re-ID systems. The MGN architecture is designed to learn global and local features independently, with each branch dedicated to a specific level of granularity. The global branch captures integral but coarse features, while the local branches focus on more detailed and fine-grained representations. The network employs softmax loss for classification and triplet loss for metric learning, ensuring robust and discriminative feature representations. Extensive experiments on mainstream datasets such as Market-1501, DukeMTMC-reID, and CUHK03 demonstrate that MGN achieves state-of-the-art performance, outperforming existing methods by a significant margin. The method's effectiveness is further validated through ablation studies, which show that the multi-branch architecture and the use of triplet loss are crucial for achieving superior results.