Cookbook running the jupyter notebook (on Windows)

[hilsonp]

Hello,
I just tried to run the jupyter notebook but, at first, it didn't seem to work as expected.

I'm running this on Windows to train on my own data.

Here is what I did:

• pip install git+https://github.com/haverland/collectmeterdigits
• pip install tensorflow
• python3 -m collectmeterdigits --collect=solar_meter --days=3
• I did sort out all images as asked
• git clone haverland/Tenth-of-step-of-a-meter-digit
• add my curated images in the collected/powermeter/emeter_c26 folder
• pip install notebook
• pip install opencv-python # Needed to avoid errors in the notebook
• pip install tensorflow-model-optimization # Needed to avoid errors in the notebook
• jupyter notebook dig_class100-s2.ipynb
• click the play button until the end (or click the "double right arrow" to play it all at once).

It takes a while and while the notebook is running, the browser tab will show an hourglass.

At the end, I took the tflite generated in the output folder, uploaded it on my ai-on-the-edge device, changed the config to use it.

The result is awesome !!! This is so accurate !

Note: 2 things helped a lot as well:

• Using a glue stick as light diffuser for the bright white led.
• Ajust the position to avoid any need to rotate the image.