[KeyPoints] - add from_mediapipe connector allowing Supervision to be used with Google MediaPipe models

[SkalskiP]

Description

• Add Skeleton.GHUM to the Skeleton enum. This is the skeleton format used by Mediapipe.
• Extend the KeyPoints object and add the from_keypoints connector.

Links

• Google Mediapipe repository: https://github.com/google/mediapipe
• Google Mediapipe example notebook: https://colab.research.google.com/github/kevinash/awesome-ai/blob/main/notebooks/6_PosesAndAction/Pose_MediaPipe.ipynb
• Google Medipipe documentation: https://developers.google.com/mediapipe/solutions/vision/pose_landmarker/

Additional

• Note: Please share a Google Colab with minimal code to test the new feature. We know it's additional work, but it will speed up the review process. The reviewer must test each change. Setting up a local environment to do this is time-consuming. Please ensure that Google Colab can be accessed without any issues (make it public). Thank you! 🙏🏻

[LinasKo]

Last time I worked on it, Mediapipe provided visiblity and not confidence. Should the implementer treat this as confidence for now?
Docs on: https://developers.google.com/mediapipe

There's cases when we have high confidence of invisible points (e.g. left or right shoulder when person is facing sideways).

(I'm not claiming the issue - just asking for details)

[LinasKo]

Let's not get bogged down by details. I propose we implement visibility as confidence for now and adjust later if required. Other than that, this is one of the easier features - get frames+keypoints from mediapipe, and implement something similar to from_ultralytics 🙂

[LinasKo]

Hey @David-rn,

I saw your comment on the RLE ticket. If you wish, this one is up for grabs. Should be pretty straightforward too.

Is this something you'd be interest in helping out with?

[David-rn]

Hi @LinasKo! I haven't worked with MediaPipe but I'll give it a try! 👍