UP-Person introduces a unified parameter-efficient transfer learning approach that enhances text-based person retrieval while drastically reducing training costs compared to full fine-tuning methods.

• Achieves superior performance with only 0.1%-1% of trainable parameters
• Introduces a unified adapter framework that works across multiple person retrieval datasets
• Provides significant computation savings while maintaining or improving accuracy
• Demonstrates practical applications for security and surveillance systems

This research matters for security operations by enabling more efficient deployment of person identification systems with minimal computational resources, making advanced surveillance technology more accessible and cost-effective for real-world security applications.

UP-Person: Unified Parameter-Efficient Transfer Learning for Text-based Person Retrieval