michalfaber / tensorflow_Realtime_Multi-Person_Pose_Estimation

Multi-Person Pose Estimation project for Tensorflow 2.0 with a small and fast model based on MobilenetV3
Other
215 stars 65 forks source link

Identify particular elements (like wrist, shoulder, knee...) #6

Open JivanRoquet opened 4 years ago

JivanRoquet commented 4 years ago

This library is really great work.

I was wondering, as it's not clear in the documentation, if there is a way to "simply" get the coordinates for the estimated location of different elements of the reconstituted pose mesh.

For example, let's say I'd like to get the following result:

{
    'wrist_R': (12, 22),
    'elbow_R': (43, 13),
    'shoulder_R': (44, 55),
    'neck': (56, 53)
}

Following through the steps in pose_mobile_net.ipynb, it doesn't like there's an immediate and/or obvious way to get to that sort of result.

In my use case, I'm actually not that interested in the "connections" between the dots, but much more in knowing what part of the body each particular dot refers to. Would you have idea of how to achieve that?

Thanks,

michalfaber commented 4 years ago

Hi @JivanRoquet I added a new folder estimation_example. The code does exactly what you need. Check out the scripts: coordinates.py, connections.py.