In this work, we present a comparison between using different pedestrian re-identification (re-id) architectures. We have investigated the advantages of using more complex and deeper convolutional neural networks (CNNs) at the feature extraction stage. The re-id network is based on the summary network presented by (Ahmed and Marks 2015) which we have modified and enhanced. The comparison is done by replacing the feature extraction portion of the network. The newer improved models performed better than the baseline model and resulted in an accuracy of above 96 on our dataset and an accuracy of 92.09 on CUHK03 test dataset. The network takes 2 images as input and, outputs a confidence level indicating whether or not the 2 images depict the same person. The 2 images both go through a CNN with shared weights and the resulting 2 feature maps are used to compare and classify the 2 images as a positive or a negative match. © 2022, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.