Person re-identification is a pre-cursor for various applications like surveillance in smart cities and high-security areas where object tracking between camera views is crucial. Recent applications in retail analytics have also been found. It is an efficient method to collect data on and track the purchasing behavior of an offline customer in a physical store. But the models come with a steep cost of a labeled dataset. The cost of a labeled dataset can be solved naturally by applying unsupervised person re-identification techniques. In the past, various architectures have been proposed for this application, for example: cluster contrast, knowledge distillation, and body part-based re-identification, etc. In this paper, some of these architectures and approaches have been implemented, enhanced, and analyzed to discuss the current challenges in the field and future work. We have fine-tuned the Teacher-Student Network to incorporate semantic information using a non-parametric Graph Convolutional Network (GCN) and the Part-Based Convolutional Neural Networks (CNNs) to include a multi-loss function. We have fine-tuned the frameworks by using different backbones and approaches. We have also provided our analysis and discussion on these results and have detailed our future work.