Pedestrian baseline re-identification method based on hierarchical self-attention network

The invention provides a pedestrian baseline re-identification method based on a hierarchical self-attention network, and belongs to the field of computer vision. According to the method, the Swin Transform is creatively used as a backbone network to be introduced into the pedestrian re-identification field, the form of the weighted sum of the ID loss and the Circle loss is used as a loss function, and through effective data preprocessing and reasonable parameter adjustment, the feature extraction capability is improved while it is ensured that the structure is simple. Compared with a traditional ResNet-based baseline method, the pedestrian baseline re-identification method has the advantage that the pedestrian re-identification effect is remarkably improved. 本发明提供了一种基于层次化自注意力网络的行人重识别基线方法，属于计算机视觉领域。本发明中，创造性的将Swin Transformer作为主干网络引入行人重识别领域，并将ID损失和Circle损失加权和的形式作为损失函数，通过有效的数据预处理和合理的调参，在保证结构简单的同时提升了特征提取能力。本发明相对于传统的基于ResNet的基线方法，显著地提高了行人重识别的效果。