This paper addresses the challenge of person re-identification (re-ID) by proposing a siamese network that combines the strengths of verification and identification models. The network simultaneously learns a discriminative embedding and a similarity measurement, improving the accuracy of pedestrian retrieval. The proposed method is applied to two large-scale re-ID datasets (Market1501 and CUHK03) and one instance retrieval dataset (Oxford5k). The results show that the proposed model outperforms state-of-the-art methods, demonstrating its effectiveness in both re-ID and instance retrieval tasks. The key contributions include the integration of verification and identification losses, the use of pre-trained CNNs, and the introduction of a non-parametric Square Layer for similarity estimation. The paper also discusses the advantages and limitations of verification and identification models, highlighting the complementary nature of the proposed approach.This paper addresses the challenge of person re-identification (re-ID) by proposing a siamese network that combines the strengths of verification and identification models. The network simultaneously learns a discriminative embedding and a similarity measurement, improving the accuracy of pedestrian retrieval. The proposed method is applied to two large-scale re-ID datasets (Market1501 and CUHK03) and one instance retrieval dataset (Oxford5k). The results show that the proposed model outperforms state-of-the-art methods, demonstrating its effectiveness in both re-ID and instance retrieval tasks. The key contributions include the integration of verification and identification losses, the use of pre-trained CNNs, and the introduction of a non-parametric Square Layer for similarity estimation. The paper also discusses the advantages and limitations of verification and identification models, highlighting the complementary nature of the proposed approach.