0

How can we modify mediapipe code to get real time gesture recognition of a basic set of gestures. I ran multi hand tracking apk but it could not recognise basic gestures as promised in https://ai.googleblog.com/2019/08/on-device-real-time-hand-tracking-with.html .

Apoorv Pandey
  • 11
  • 1
  • 3

1 Answers1

0

So far, I see two ways:

  1. Use a pre-trained model wrapped within a calculator.
  2. Train a model using the 21 landmarks by collecting points for each gesture and normalizing the landmarks' x, y, z values.
prantoran
  • 1
  • 2
  • i am using their landmark model, i input the image and got output of shape (1,63), can you tell me what this is and how to use them to train model – Ankush Rawat Jun 24 '20 at 13:13
  • 1
    I checked out the examples for Linux desktop using C++. What I did was identify the possitions of the landmarks with respect to their locations in palms and manually wrote detectors for gestures by calculating presence of polygon shapes. – prantoran Jun 25 '20 at 08:20
  • https://github.com/google/mediapipe/issues/414 i found this link useful for my particular problem hope it will be useful for other too – Ankush Rawat Jun 25 '20 at 09:48