This paper challenges the prevailing belief that triplet loss is inferior to surrogate losses like classification or verification for person re-identification (ReID). Instead, it shows that a variant of the triplet loss can outperform existing methods on three major ReID datasets: CUHK03, Market-1501, and MARS. The key contribution is the introduction of a batch hard loss, which eliminates the need for offline hard negative mining and achieves state-of-the-art results both with pretrained and scratch-trained networks. The paper also demonstrates that training from scratch can yield competitive results, challenging the notion that pretrained models are essential for ReID. The triplet loss is shown to be effective for learning discriminative embeddings, with the batch hard variant performing best. The paper also discusses the importance of proper training strategies, the impact of input size, and the benefits of end-to-end learning. Overall, the study highlights the effectiveness of the triplet loss in ReID and its potential to significantly impact future research in the field.This paper challenges the prevailing belief that triplet loss is inferior to surrogate losses like classification or verification for person re-identification (ReID). Instead, it shows that a variant of the triplet loss can outperform existing methods on three major ReID datasets: CUHK03, Market-1501, and MARS. The key contribution is the introduction of a batch hard loss, which eliminates the need for offline hard negative mining and achieves state-of-the-art results both with pretrained and scratch-trained networks. The paper also demonstrates that training from scratch can yield competitive results, challenging the notion that pretrained models are essential for ReID. The triplet loss is shown to be effective for learning discriminative embeddings, with the batch hard variant performing best. The paper also discusses the importance of proper training strategies, the impact of input size, and the benefits of end-to-end learning. Overall, the study highlights the effectiveness of the triplet loss in ReID and its potential to significantly impact future research in the field.