Clarification on the world-landmark-coordinates of the Mediapipe Hand-Landmarker

TensorFlow General Discussion
1

Background
To make some finger-related measurements (in centimeters) without a reference object, I am planning to use the Mediapipe Hand-Landmarker model to detect specific landmarks and then use the hand_world_landmarks property of the HandLandmarkerResult class to calculate the pixel-per-metric value. I wasn’t sure if this would work, but I tested this out on a couple of images and it seemed to work.

Question
Now, I want to extend this to measure the finger depths and I was wondering if the z axis value of hand_world_landmarks would be a good estimator for this? If not, are there any alternate ways I can achieve this (using Mediapipe)?

3

Hi @Gokulakrishnan_Shank ,

Yes the z axis of the `hand_world_landmarks` can be used to estimate relative depths, but the 21 landmarks provided by hands are primarily skeletal landmarks, they do not represent the surface or thickness of the fleshy parts of the finger.

Currently there are no direct alternatives within hands landmarker API to get the finger thickness measurements.

Thank You