TensorFlow Lite Model Maker issues

General Discussion
,
#1

Hi, I’m currently trying to train a TensorFlow Lite Model for Object Detection using Raspberry Pi. To do so, I’m using the “Model Maker Object Detection for Android Figurine” Colab notebook provided by TensorFlow (Google Colab).

I have already seen that many people had issues using this notebook in the past month, as there is a compatibility problem with Python 3.10.* Some of them were able to solve this problem by rolling back to Python 3.9.* (see Bug - pip install tflite-model-maker fails or hangs forever · Issue #60431 · tensorflow/tensorflow · GitHub) . Unfortunatly this roll back option expired mid-May.

I tried to get back to Python3.9.* using:
!sudo update-alternatives --install /usr/bin/python3 python3 /usr/bin/python3.9 2
Then I was able to install tflite-model-maker, but there were hundreds of dependencies which I couldn’t figure out.

I would be so grateful if anyone could help me to figure out a solution for this problem.

Best landoVdB

1 Like
#3

@landoVdB,

Welcome to the Tensorflow Forum!

This is known issue and engineering team are working on it.

You can try to use the MediaPipe Model Maker

For more details, please refer to the Object detection model customization guide may help you.

Thank you!

#4

@chunduriv Thank you for your fast reply.
Is it possible to use the models trained with MediaPipe Model Maker on a Raspberry Pi?

#6

@landoVdB,

I hope it should work, but i have never tried it.

Thank you!