{
  "paper_id": "53a1e6963a229942572ffd06a16f7a06",
  "paper_title": "AlphaPose: Whole-Body Regional Multi-Person Pose Estimation and Tracking in Real-Time",
  "paper_authors": [
    "Hao-Shu Fang",
    "Jiefeng Li",
    "Hongyang Tang",
    "Chao Xu",
    "Haoyi Zhu",
    "Yuliang Xiu",
    "Yong-Lu Li",
    "Cewu Lu"
  ],
  "publication_date": "",
  "publication_venue": "IEEE Transactions on Pattern Analysis and Machine Intelligence",
  "abstract": "Accurate whole-body multi-person pose estimation and tracking is an important yet challenging topic in computer vision. To capture the subtle actions of humans for complex behavior analysis, whole-body pose estimation including the face, body, hand and foot is essential over conventional body-only pose estimation. In this article, we present AlphaPose, a system that can perform accurate whole-body pose estimation and tracking jointly while running in realtime. To this end, we propose several new techniques: Symmetric Integral Keypoint Regression (SIKR) for fast and fine localization, Parametric Pose Non-Maximum-Suppression (P-NMS) for eliminating redundant human detections and Pose Aware Identity Embedding for jointly pose estimation and tracking. During training, we resort to Part-Guided Proposal Generator (PGPG) and multi-domain knowledge distillation to further improve the accuracy. Our method is able to localize whole-body keypoints accurately and tracks humans simultaneously given inaccurate bounding boxes and redundant detections. We show a significant improvement over current state-of-the-art methods in both speed and accuracy on COCO-wholebody, COCO, PoseTrack, and our proposed Halpe-FullBody pose estimation dataset. Our model, source codes and datasets are made publicly available at <https://github.com/MVIG-SJTU/AlphaPose>.",
  "keywords": [
    "Human pose estimation",
    "pose tracking",
    "whole-body pose estimation",
    "hand pose estimation",
    "realtime",
    "multi-person"
  ],
  "paper_institutions": [
    "Shanghai Jiao Tong University"
  ],
  "output_language": "zh",
  "provider": "doubao",
  "model": "ark-code-latest",
  "prompt_template": "simple",
  "extracted_at": "2026-08-16T19:46:35.633301Z",
  "template_tag": "simple",
  "summary": "### 摘要\n\n本文提出AlphaPose系统，针对多人物全身体态估计与跟踪这一挑战。现有方法在检测阶段与姿态估计分离，存在漏检和速度慢的问题；同时，热图表示存在量化误差，对脸部和手部等精细关键点定位不准确。AlphaPose引入对称积分关键点回归（SIKR）以消除量化误差，采用参数化姿态非极大值抑制（P-NMS）处理冗余检测，并利用姿态感知身份嵌入实现联合估计与跟踪。训练中采用部分引导的候选框生成器（PGPG）和多域知识蒸馏以提升泛化能力。在COCO-wholebody、COCO、PoseTrack及新标注的Halpe-FullBody数据集上，AlphaPose在速度与精度上均超越现有方法，并支持实时运行。实验表明，SIKR显著提高脸部和手部定位精度，PGPG解决域差异，跟踪模块在PoseTrack上取得高MOTA且速度较快。未来可扩展至3D关键点和网格估计，适用于人机交互、电影制作和动作识别等场景。核心方法包括对称梯度设计和两步热图归一化。\n\n### 关键要点\n\n- 提出AlphaPose系统，统一实现全身体态估计与跟踪，并支持实时运行。\n- 引入对称积分关键点回归（SIKR），解决热图量化误差，首次在回归方法上达到与热图相当精度。\n- 提出参数化姿态NMS（P-NMS），通过数据驱动优化消除冗余检测，降低对强检测器的依赖。\n- 采用姿态感知身份嵌入，联合进行姿态估计和跟踪，并利用姿态引导注意力机制提升重识别特征质量。\n- 提出部分引导的候选框生成器（PGPG），模拟检测器输出分布，增强训练数据多样性。\n- 采用多域知识蒸馏，整合COCO-WholeBody、300Wface、FreiHand等数据集，提升手部和脸部关键点精度。\n- 标注Halpe-FullBody数据集，包含136个关键点，覆盖身体、脚、手和脸，推动全身体态研究。\n- 在Halpe-FullBody上，FastPose50-si较热图版本提高2.4 mAP，提升主要来自脸和手。\n- 在COCO-WholeBody上，FastPose以较小输入尺寸和计算量取得最高精度，如FastPose50-dcn-si达到57.7 AP。\n- 在PoseTrack上，Ours-SEP取得76.9 mAP和65.7 MOTA，速度达8.9 fps，优于多数方法。\n- 消融实验验证了两步热图归一化、SIKR、P-NMS和PGPG的贡献，分别提升精度。\n- AlphaPose库支持多种检测器和姿态估计器，采用五阶段并行流水线，实现实时推断。"
}