PosePipe: Open-Source Human Pose Estimation Pipeline for Clinical Research

PosePipe is a human pose estimation (HPE) pipeline designed to facilitate movement analysis from videos.
It uses DataJoint to manage relationships between algorithms, videos, and intermediate outputs.

Key features:

• Modular wrappers for numerous state-of-the-art HPE algorithms
• Structured video and data management via DataJoint
• Output visualizations to easily compare and analyze results
• Designed for clinical research movement analysis pipelines

Quick Start

1. Install PosePipe

pip install pose_pipeline

Detailed installation instructions are provided to launch a DataJoint MySQL database and install OpenMMLab packages.

2. Test the pipeline

Use the Getting Started Notebook to start running your videos through the pose estimation framework.

Recent Updates and Supported Algorithms

• Upgraded mmcv to v2.x
• Tracking Algorithms (from mmdetection):
  • DeepSORT
  • QDTrack
• Top Down 2D Body Keypoint Detection Algorithms (from mmpose):
  • RTMPose + RTMPose on Coco-Wholebody
  • RTMPose + RTMW on Cocktail14
  • Topdown Heatmap + ViTPose on Coco
• Top Down 2D Hand Keypoint Detection Algorithms (from mmpose):
  • RTMPose + RTMPose on Hand5
  • RTMPose + RTMPose + Coco + Wholebody + Hand on Coco_wholebody_hand
  • Topdown Heatmap + Resnet on Freihand2d
  • Topdown Heatmap + HRNetv2 + Dark on Rhd2d
  • Topdown Heatmap + HRNetv2 + UDP on OneHand10k
• Bottom Up Algorithms:
  • MeTRAbs

Developer Setup

VSCode is recommended for development.

Include the following in your .vscode/settings.json to enable consistent black formatting:

{
  "python.formatting.blackArgs": [
    "--line-length=120",
    "--include='*py'",
    "--exclude='*ipynb'",
    "--extend-exclude='.env'",
    "--extend-exclude='3rdparty/*'"
  ],
  "editor.rulers": [120]
}

Project Info

• License: GPL-3.0
• Source Code: GitHub Repo
• PyPI: https://pypi.org/project/posepipe
• Issues/Contributions: Please use Issues for bug reports and feature requests

Citation

If you use this tool for research, please cite:

@misc{posepipe2024,
  author       = {R James Cotton},
  title        = {PosePipe: Open-Source Human Pose Estimation Pipeline for Clinical Research},
  year         = {2024},
  howpublished = {\url{https://github.com/IntelligentSensingAndRehabilitation/PosePipeline}}
}