Poses are detected with the PoseNet Model using TensorFlow.js

Due to WebGL precision problems, TensorFlow.js works less accurate in Safari. Currently, best results are obtained with Firefox or Chrome.

The applet shows an IFS with two similarities based (each mapping the segment shoulder-elbow to the trunk) on the pose of the pose of a person. A Feedback loop-approach renders the IFS in CindyGL.

Idea from MoMath