This repository serves as a comprehensive survey and curated list of academic papers focused on Pedestrian Attribute Recognition (PAR) and related computer vision tasks. It aims to provide researchers and practitioners with an up-to-date overview of the field, including key datasets, methodologies, and recent advancements, thereby facilitating research and development in human-centric perception.

How It Works

The project compiles and categorizes research papers, offering insights into the evolution of PAR techniques. It highlights performance comparisons on benchmark datasets like PETA and RAP, noting trends such as the significant outperformance of deep learning methods over traditional approaches and the recent plateau in accuracy improvements. The list also tracks emerging datasets, challenges, and relevant open-source toolkits, providing a structured view of the research landscape.

Quick Start & Requirements

This is a curated list of academic papers and not a software project. Therefore, there are no installation or execution requirements.

Highlighted Details

• Features performance comparisons on PETA and RAP datasets, showing deep learning methods outperforming older techniques.
• Discusses the potential bottleneck in current deep learning-based PAR algorithms.
• Tracks recent developments including new datasets (EventPAR, MSP60K), contests (PAR CONTEST 2025, WACV'24), and talks.
• Provides links to related resources such as the OpenPAR toolkit and numerous datasets and code repositories for specific papers.
• Organizes papers by year and related tasks like Person Re-ID, Pedestrian Detection, and Action Recognition.

Maintenance & Community

The repository is maintained by Xiao Wang, with contact provided via email (wangxiaocvpr@foxmail.com) for paper submissions. A WeChat group is available for community discussion. Users are encouraged to cite the primary survey paper if the resource is found useful.

Licensing & Compatibility

No specific software license is mentioned for the repository or the curated list itself.

Limitations & Caveats

As a paper list, it does not offer a runnable system or code. The content reflects the state of research up to the last update mentioned (April 2025), and new publications may not be immediately included. The "How It Works" section describes the research field's trends rather than a specific technical implementation.