2

Is there a way to selectively mask pose landmarks in MediaPipe? MediaPipe allows me to apply a segmentation mask to a pose with the enable_segmentation parameter, but it gives me the entire body. Is it possible to, for example, segment only the right leg? The left hand?

Pose detection:

Example animated image of MediaPipe detecting body landmarks of a man doing yoga

Segmentation Mask:

Image of a man dancing. The background is washed out with the help of a MediaPipe pose segmentation mask.

Ian Hunter
  • 9,466
  • 12
  • 61
  • 77
  • https://google.github.io/mediapipe/solutions/pose.html#pose_landmarks Why can't you use the pose points to pick the body parts. For example 26,28,30,32 gives the right leg. – Vijay Mariappan Oct 27 '22 at 15:24
  • 1
    pose estimation is very much different from instance/semantic segmentation. if the segmentation network wasn't trained for this flexibility, you'll have a task ahead of you. – Christoph Rackwitz Oct 27 '22 at 21:57

0 Answers0