Pixel coordinates of each hand keypoint for hand detection

[chillcloud-dev]

Hi MediaPipe team

i am working on hand gesture recognition and for my project I need the z - coordinates of each hand.
I extract the z - coordinates by changing the subgraph hand_landmark_gpu.pbtxt
from model_path: "mediapipe/models/hand_landmark.tflite" -----> model_path: "mediapipe/models/hand_landmark_3d.tflite"

and by applying the solution from #64, #99 and #159 I am able to extract the x, y and z fir each hand in each frame.

So far so good !

But, the z- coordinate is pretty much useless since (according to link) , the z-coordinate is normailzed relative to the wrist. And it doesn't have any unit and is in a range between +- 60 as far as i can tell. So there is no relevance of having this dimensional information.

Therefore, I want to use the depth channel of my realsense camera D415 to get the depth information by using the pixel coordinates of x, y of each key point.

Could you tell me, where I can extract the pixel coordinates of each key point in the camera frame ?
Since mediapipe outputs the x, y coordinates is it possible to get also the pixel coordinates of each keypoint ?

Thanks :)

best regards

[mahdinobar]

+1

[fanzhanggoogle]

For example, you get the coordinates here and simply multiply by the input image size?

[chillcloud-dev]

@fanzhanggoogle
already done that 👍

its the easiest way