Generative Models for Person Re-identification

Person re-identification is a technique used in AI and machine learning to recognize a person across different images or videos — even with different angles, lighting, or clothing. It's used in security surveillance, digital image identification, and video behavior analysis.

The main challenge is the variability in appearance (clothing, hairstyle) and the limited size of available training datasets due to privacy concerns. The best public datasets are limited: Market1501 (1,501 people, 6 cameras) and DukeMTMC-reID (702 people, 8 cameras).

This project investigates the use of Generative Adversarial Networks (GANs), combined with data augmentation techniques, to generate synthetic images that increase the quantity and diversity of training data for re-identification models.