Sunday 09 March 2025
A new system for identifying individuals in crowded spaces, without the need for labelled data or human supervision, has been developed by researchers. The approach uses a combination of visual transformers and memory banks to learn discriminative features that can be used to identify people from surveillance footage.
The problem of person re-identification is a challenging one, particularly when it comes to identifying individuals in crowded areas like public spaces, train stations or airports. Traditional methods rely on manually labelling data, which can be time-consuming and expensive, or using supervised learning approaches that require large amounts of labelled data. However, these methods often struggle to generalise well to new environments or populations.
The new system, called TCMM for short, tackles this problem by using a combination of visual transformers and memory banks. Visual transformers are a type of neural network that is particularly effective at processing images and extracting relevant features. They work by dividing an image into smaller patches and then processing each patch separately before combining the results to form a complete representation of the image.
The memory bank component of TCMM is used to store a set of prototypes, or example images, that are representative of different individuals. These prototypes are used as positive examples during training, helping the network to learn what characteristics make one person distinct from another.
During testing, the system uses a combination of instance and prototype-level memory modules to identify new individuals. The instance memory module is used to store information about specific individuals, while the prototype memory module is used to retrieve prototypes that match the features extracted from the test image.
The results of the study are promising, with TCMM achieving state-of-the-art performance on two widely-used benchmarks for person re-identification. The system was also able to generalise well to new environments and populations, demonstrating its potential for real-world applications.
One of the key advantages of TCMM is its ability to learn from unlabelled data, making it a more practical solution than traditional supervised learning approaches. This could be particularly useful in situations where labelled data is not available or is difficult to obtain.
The development of TCMM has the potential to improve our ability to identify individuals in crowded spaces, and could have applications in areas such as surveillance, security and law enforcement. However, further research is needed to fully understand the capabilities and limitations of this new system.
Cite this article: “Person Re-Identification System Learns from Unlabelled Data”, The Science Archive, 2025.
Person Re-Identification, Surveillance, Visual Transformers, Memory Banks, Neural Networks, Image Processing, Instance Memory Module, Prototype Memory Module, Unlabelled Data, Security
