Person re-identification with transformers and image stacking

Person re-identification (ReID) is increasingly important due to the expansion of surveillance cameras. ReID can effectively operate in various conditions, making it suitable for security, retail analytics, and smart city applications. We propose a transformer-based model, DeepChangeVIT-ReID, fine-tuned with triplet loss, using the DeepChange dataset. We address long-term ReID challenges, including pose variations, camera angle differences, and clothing alterations. DeepChangeVIT-ReID achieves state-of-the-art performance, significantly improving Rank-1 accuracy compared to existing methods on the DeepChange dataset.