Research on Person Re-Identification Method Based on Fine-tune ResNet50 Network
Xiuhua Hu, Yuan Chen, Xinyu Ma, Yingyu Liang · 2020
Aiming at the problem that the unreasonable design of the traditional network structure may easily affect the performance of person re-identification, this paper provides a method for person re-identification based on fine-tuning ResNet50 network. This method builds a model based on the ResNet50 network structure. In the model training stage, the data set is preprocessed, the data enhancement method is adopted to avoid over-fitting, select the appropriate loss function abate the influence of the gradient disappearance, and use the cosine distance measurement function to complete the spatial distance calculation and similarity ranking between sample. The experimental results show that the designed new method can improve the performance of person re-identification to a certain extent.