Camera-aware Label Refinement for Unsupervised Person Re-identification

2024-03-25

arXiv.org

AI Summary

To address severe pseudo-label noise and performance degradation in unsupervised person re-identification (ReID) caused by inter-camera feature distribution shifts, this paper proposes a camera-aware label refinement framework. Methodologically: (1) it introduces intra-camera clustering for pseudo-label generation—first achieving camera-specific clustering to suppress local noise; (2) it designs a self-paced global label refinement strategy that progressively enforces cross-camera label consistency; and (3) it incorporates a feature-level camera alignment module to mitigate inter-domain feature shift. Extensive experiments demonstrate significant improvements over state-of-the-art methods on Market-1501 and DukeMTMC-reID, validating the effectiveness of leveraging camera priors in unsupervised ReID modeling. The source code is publicly available.

Technology Category

• Computer Vision: Motion & Tracking
• Machine Learning: Semi-Supervised Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Unsupervised person re-identification aims to retrieve images of a specified person without identity labels. Many recent unsupervised Re-ID approaches adopt clustering-based methods to measure cross-camera feature similarity to roughly divide images into clusters. They ignore the feature distribution discrepancy induced by camera domain gap, resulting in the unavoidable performance degradation. Camera information is usually available, and the feature distribution in the single camera usually focuses more on the appearance of the individual and has less intra-identity variance. Inspired by the observation, we introduce a extbf{C}amera- extbf{A}ware extbf{L}abel extbf{R}efinement~(CALR) framework that reduces camera discrepancy by clustering intra-camera similarity. Specifically, we employ intra-camera training to obtain reliable local pseudo labels within each camera, and then refine global labels generated by inter-camera clustering and train the discriminative model using more reliable global pseudo labels in a self-paced manner. Meanwhile, we develop a camera-alignment module to align feature distributions under different cameras, which could help deal with the camera variance further. Extensive experiments validate the superiority of our proposed method over state-of-the-art approaches. The code is accessible at https://github.com/leeBooMla/CALR.

Problem

• Reduces camera discrepancy in unsupervised person re-identification
• Improves clustering accuracy by refining intra-camera pseudo labels
• Aligns feature distributions across different camera domains

Innovation

• Intra-camera training for reliable pseudo labels
• Camera-alignment module reduces feature distribution discrepancy
• Self-paced learning with refined global pseudo labels