adirik / lightweight-openpose

PyTorch version of Lightweight OpenPose as introduced in "Real-time 2D Multi-Person Pose Estimation on CPU: Lightweight OpenPose"

  • Public
  • 1.6K runs
  • GitHub
  • Paper
  • License

Lightweight OpenPose

Demo for Real-time 2D Multi-Person Pose Estimation on CPU: Lightweight OpenPose as implemented in the original repository. The available weights are downloaded from the link provided at the Github repository of this project.

API Usage

The API returns the 18 human body keypoints per person in the input image. The output data is as follows:

  • keypoint_names: Ordered list of 18 keypoint names for the human body
  • json_data: A dictionary containing the id, confidence, bbox (x1, y1, width, height), keypoints (list of all keypoints - [x, y] pairs) of each detected person.
    • Invisible / undetectable keypoints are returned as [-1, -1].
  • keypoints_img (optional): Input image overlayed with detected skeletons.

References

@inproceedings{osokin2018lightweight_openpose,
    author={Osokin, Daniil},
    title={Real-time 2D Multi-Person Pose Estimation on CPU: Lightweight OpenPose},
    booktitle = {arXiv preprint arXiv:1811.12004},
    year = {2018}
}