1

I am using PoseNet from TensorFlow Lite library to detect key point on images. What is the best way to evaluate the results. I feed in images and get key points as a result. How can I say if the result is good, bad or average?

Parth
  • 35
  • 8

1 Answers1

0

From the output, you can extract the confidence score for each point, between 0.0 and 1.0. You can then determine whether the average confidence score is greater than a set threshold (e.g., 0.5) to tell whether the result is good or bad.

Please see the example apps to learn how the confidence score can be obtained.

Android example: https://github.com/tensorflow/examples/blob/master/lite/examples/posenet/android/posenet/src/main/java/org/tensorflow/lite/examples/posenet/lib/Posenet.kt#L259

iOS example: https://github.com/tensorflow/examples/blob/master/lite/examples/posenet/ios/PoseNet/ModelDataHandler/ModelDataHandler.swift#L229

yyoon
  • 3,555
  • 3
  • 21
  • 17
  • It's still detecting the keypoints incorrectly after I set a threshold of the confidence score. I wanted a mechanism where I could test the quality towards the end when I already have the results from PoseNet. – Parth Oct 23 '20 at 08:44