This research evaluates the capabilities of Large Vision-Language Models (LVLMs) in identifying and tracking individuals across different camera views.

• Compares performance of leading LVLMs in human re-identification tasks
• Assesses limitations of general-purpose AI models in specialized security contexts
• Identifies the gap between LVLMs and state-of-the-art domain-specific solutions
• Provides insights for security professionals considering LVLM implementation

For security applications, this research delivers crucial benchmarks that help organizations make informed decisions about AI integration in surveillance systems, highlighting where specialized solutions still outperform general AI capabilities.

Human Re-ID Meets LVLMs: What can we expect?