# Person Re-Identification by Camera Correlation Aware Feature Augmentation

**Authors:** Ying-Cong Chen, Xiatian Zhu, Wei-Shi Zheng, Jian-Huang Lai  
**arXiv:** [1703.08837](https://arxiv.org/abs/1703.08837)  
**Date:** 2017-03-26  
**Summary page:** https://www.summarizepaper.com/en/paper/1703.08837v1/

## Key points

- Authors: Ying-Cong Chen, Xiatian Zhu, Wei-Shi Zheng, Jian-Huang Lai
- Framework: Camera coRrelation Aware Feature augmenTation (CRAFT)
- Addresses challenge of person re-identification in multi-camera environments
- CRAFT operates from a feature augmentation perspective
- Measures camera correlation to transform original features into a new adaptive space
- Enhances model effectiveness by considering view-specific characteristics
- Can be extended to learn view-specific feature transformations across networks with more than two cameras
- Introduces domain-generic deep person appearance representation designed to be view-invariant
- Offers comprehensive solution for person re-identification complexities in multi-camera environments

## Summary

In their paper "Person Re-Identification by Camera Correlation Aware Feature Augmentation," authors Ying-Cong Chen, Xiatian Zhu, Wei-Shi Zheng, and Jian-Huang Lai present a novel approach to the challenge of person re-identification (re-id) in multi-camera environments. Their proposed framework, called Camera coRrelation Aware Feature augmenTation (CRAFT), addresses the significant and unknown cross-view feature distortion that complicates this task. CRAFT operates from a feature augmentation perspective and adaptively measures camera correlation from cross-view visual data distribution to transform original features into a new adaptive space. This allows for more accurate representation of view-specific characteristics and enhances the overall effectiveness of the model by considering both general and specific aspects of each camera view. Additionally, CRAFT can be extended to jointly learn view-specific feature transformations across networks with more than two cameras. The authors also introduce a domain-generic deep person appearance representation designed to be view-invariant, which contributes to improved performance in person re-identification tasks. Overall, CRAFT offers a comprehensive solution for addressing the complexities of person re-identification through tailored feature augmentation and adaptive transformation techniques specifically designed for multi-camera environments.