/LabelArticulation

an articulated object annotation tool

Primary LanguagePythonMIT LicenseMIT

LabelArticulation

LabelArticulation is an interactive articulated object modeling tool that implements:

  • Align origin and axis of raw scanned point cloud
  • Interactive part segmentation
  • Joint annotation (joint location and direction)
  • URDF file export

We also support rigid and articulated object 6D pose labeling from captured RGB-D images

Requirements

Installation

$ pip install -r requirements.txt

Update

  • 4/2021: Add rigid and articulated object 6D pose labeling tools

License

LabelArticulation is released under the MIT license.

Citing LabelArticulation

If you use LabelArticulation in your research to build dataset, please use the following BibTeX entry.

@misc{LabelArticulation2021,
  author =       {Liu Liu and Han Xue and Wenqiang Xu},
  title =        {LabelArticulation},
  howpublished = {\url{https://github.com/liuliu66/LabelArticulation}},
  year =         {2021}
}

Articulation object modeling

After installation, please see Tutorial (articulation modeling) for brief tutorials on how to model the raw scanned point cloud into articulation.

6D pose annotation

please see Tutorial (articulation pose label) for brief tutorials on how to label 6D pose of articulated objects.

How to get RGB and depth image?

We strongly recommend to follow pyrs repository. Images of directory "data/" were captured using that repository.

ToDo

  • output a total transformation matrix
  • add input arguments (model name, scene name)
  • add input argument of initial pose
  • add part-level 6D pose annotation
  • add prismatic and revolute joint annotation
  • add function to choose various models
  • visualize depth data of input scene
  • visualize coordinate axis
  • realsense to capture rgbd images
  • annotate point cloud and mesh at the same time