Adversarial Pairwise Reverse Attention for Camera Performance Imbalance in Person Re-identification: New Dataset and Metrics

TL;DR

This paper addresses camera performance imbalance in person re-identification by introducing a new dataset, metrics, and an adversarial attention module to improve camera-invariant features.

Contribution

It presents the first dataset and metrics specifically for camera imbalance, and proposes the APRA module with a pairwise attention inversion mechanism.

Findings

New dataset from 38 cameras for imbalance analysis

Novel metrics to quantify camera performance imbalance

APRA module improves camera-invariant feature learning

Abstract

Existing evaluation metrics for Person Re-Identification (Person ReID) models focus on system-wide performance. However, our studies reveal weaknesses due to the uneven data distributions among cameras and different camera properties that expose the ReID system to exploitation. In this work, we raise the long-ignored ReID problem of camera performance imbalance and collect a real-world privacy-aware dataset from 38 cameras to assist the study of the imbalance issue. We propose new metrics to quantify camera performance imbalance and further propose the Adversarial Pairwise Reverse Attention (APRA) Module to guide the model learning the camera invariant feature with a novel pairwise attention inversion mechanism.