A real-time demo running on webcam feed that guides subject to move to match pre-defined target poses.
Move your body to match each of the target poses on the right!
conda env create -f environment.yml -n movenet-dance-demo
conda activate movenet-dance-demo
To change the target poses, see the notebook: make_target_poses.ipynb
For an example of standalone inference, see the notebook: inference_demo.ipynb
To run the demo:
python live_demo.py
To use a different camera:
python live_demo.py -c 1
To make it easier, increase the target tolerance:
python live_demo.py --tolerance 0.7
Hotkeys:
1 - 9: Switch between target poses.
Tab: Cycle through target poses.
R: Reset completed poses.
Q or Esc: Quit the demo.
This demo was written by Talmo Pereira at the Salk Institute for Biological Studies to be presented at the SICB 2022 Annual Meeting.
This code uses MoveNet for pose estimation by Ronny Votel, Na Li and other contributors at Google.
The target pose data is courtesy of the danceTactics crew directed by Keith A. Thompson at Arizona State University.