TensorFlow Yoga Pose Script Doesn't Classify Accurately


I am currently trying to setup an ml tflite model for yoga pose classification using this page: (1) Human Pose Classification with MoveNet and TensorFlow Lite

I have also downloaded the already processed tflite model from here: (2) https://storage.googleapis.com/download.tensorflow.org/models/tflite/pose_classifier/yoga_classifier.tflite

After copying the TensorFlow script from link 1 and running it on some of the training images, most of the predictions it makes are incorrect. Then when running the tflite model from link 2 it seems to be almost always correct when making predictions. This doesn’t make sense to me for 2 reasons:

  1. The ML models should both be the same since the tutorial from link 1 demonstrates how to create the same yoga pose classifier I have downloaded from link 2.

  2. I have copied everything from link 1 and ran the program, then tested the tflite model and labels on a few of the training images which should simply classify them with the same label they were trained with, yet this is not the case.

Hopefully this all makes sense and you can see my problem? I am very unsure as to why this is happening and any help will be appreciated :slight_smile:

If you need any further details or have any questions about the training (copied from link 1) or pose_estimation.py test script (copied from examples/lite/examples/pose_estimation/raspberry_pi at master · tensorflow/examples · GitHub) please ask.

Thank you,

1 Like

I’m seeing the same problem. Did you resolve yours?