3DPoseViewer

A python viewer for visualizing 3D human poses and facilitating debugging of pose related tasks. This code supplements the following paper:

Proactive Multi-Camera Collaboration for 3D Human Pose Estimation (ICLR 2023)

Features

• Support multi-human and multi-camera.
• Written in python, easy to use and hack.
• Based on Qt, easy to add various interactive widgets.
• Cross-platform support: Windows, Linux and macOS.

Install

Requirements:

• python >= 3.9
• numpy
• pyqtgraph
  • PyQt5
  • pyopengl
  • cupy (optional)
• pyav
• pims

pip install -r requirements.txt

Quickstart

1. Save your 3d pose data as a numpy data file .npz.
2. Save 2D images captured from different views as seperate video files .mp4.
3. Open them with the visualizer.

python -m visualize  # open the example data under /examples/seq1

Feed Your Own Data

3D data (.npz file)

• gt3d: GT 3d human pose sequence. Numpy array of shape [t, max_Ngt, j, 3]. t: frame id. max_Ngt: max number of gt humans across the whole sequence. Fill zeros for missing humans and joints. [j, 3]: 3D location of j joints.

• pred3d: Predicted 3d human pose sequence. Numpy array of shape [t, max_Npred, j, 3]. max_Npred: max number of pred humans. Fill zeros for missing humans and joints.

• camera (optional): camera location sequence. Numpy array of shape [t, max_c, 5, 3]. max_c: max number of cameras in a frame. More detailes can be found here.

• map_center (optional): center of map, used to offset the ground plane. Default to [0, 0, 0].

2D view (optional)

• Video files, .mp4, .mov, .avi, the same length as 3D data.

Citation

If you find this viewer helpful, please cite:

@inproceedings{ci2023proactive,
  title={Proactive Multi-Camera Collaboration for 3D Human Pose Estimation},
  author={Hai Ci and Mickel Liu and Xuehai Pan and fangwei zhong and Yizhou Wang},
  booktitle={The Eleventh International Conference on Learning Representations },
  year={2023},
  url={https://openreview.net/forum?id=CPIy9TWFYBG}
}

License

Apache License, Version 2.0.