Towards Open-World Person Re-Identification by One-Shot Group-Based Verification

Area of Science:

• Computer Vision
• Machine Learning

Background:

• Person re-identification (re-id) matches individuals across non-overlapping camera views.
• Real-world scenarios often involve open-world re-id with few images per target.
• Existing methods struggle with open-world challenges due to closed-world assumptions and multi-shot learning requirements.

Purpose of the Study:

• To address the limitations of current person re-identification methods in open-world scenarios.
• To develop a novel approach for one-shot group-based verification in person re-identification.
• To leverage information from labeled non-target datasets for improved re-id performance.

Main Methods:

• Formulation of a novel transfer local relative distance comparison (t-LRDC) model.
• Implementation of one-shot group-based verification for open-world person re-identification.
• Mining and transferring information from a labeled open-world non-target dataset.

Main Results:

• The proposed t-LRDC model effectively addresses the open-world person re-identification problem.
• The approach demonstrates superior performance compared to non-transfer learning methods.
• Experimental results show outperformance against existing transfer learning-based re-id methods.

Conclusions:

• The t-LRDC model offers a significant advancement in open-world person re-identification.
• Transfer learning from non-target data is a viable strategy for improving re-id accuracy.
• The proposed method provides a robust solution for real-world person matching challenges.