基于多维度注意力机制的行人重识别算法研究

Research on pedestrain re⁃identification algorithm based on multi⁃dimensional attention mechanism

针对有监督行人重识别算法泛化能力较差的问题,提出一种基于多维度注意力机制的行人重识别算法。该算法引入了一个多维度注意力模块,通过学习行人图像在多个维度上的特征信息,提高行人图像泛化能力。此外,还提出了一个优化损失模块,加速身份损失的收敛,从而提高模型的训练速度。通过将这2个模块结合起来,本文算法在公开数据集Market—1501和DukeMTMC⁃ReID上进行了大量实验。实验结果表明:所提出的算法能够有效地解决有监督行人重识别算法泛化能力较差的问题。

Aiming at the problema of poor generalization capability of supervised pedestrain re⁃identification algorithms,a pedestrain re⁃identification algorithm based on multi⁃dimensional attention mechanism is proposed.A multi⁃dimensional attention module is introduced to improve the generalization ability of pedestrain images by learning feature information for pedestrain images in multiple dimensions.Additionally,an optimized loss module is proposed to accelerate the convergence of identity loss and improve the training speed of the model.By combining the two modules,the proposed algorithm is extensively tested on the public datasets,Market—1501 and DukeMTMC⁃ReID.The experimental results demonstrate that the proposed algorithm can effectively solve the problem of poor generalization capability of supervised pedestrain of re⁃identification.