Lightweight Unsupervised Pre-Trained Person Re-Identification Implementation

Person Re-identification (Re-ID) is a critical task in computer vision with broad applications in surveillance and security systems. Achieving high accuracy while maintaining a lightweight and computationally efficient design remains a major challenge, particularly in resource-constrained environments. In this paper, we present a novel lightweight Re-ID model that integrates an additive attention mechanism with cross-layer adaptive weighting and multi-layer feature fusion. By adopting a hierarchical design, the model effectively balances local detail extraction in shallow layers with global semantic representation in deeper layers, ensuring both efficiency and performance. Extensive experiments conducted on the Market-1501 dataset demonstrate the effectiveness of our approach, achieving a mean Average Precision (mAP) of 75.6% and a Top-1 accuracy of 90.8%. The proposed model not only outperforms several existing methods but also undergoes diverse experimental validations to further confirm its practical applicability. This approach offers an innovative solution for lightweight Re-ID, delivering efficient and accurate performance in real-world scenarios.