Conditional import of Keras/TFLite in model.py

[fegue]

When running BirdNET on a Raspberry Pi it is sufficient to install tflite-runtime without the need to install tensorflow.
Not having to install tensorflow/keras keeps the installation smaller and the process simpler.

It would be great if the import could be conditional so that the script does not throw an error when there is no installation of tensorflow.

It could be something like

if cfg.MODEL_PATH.endswith('.tflite'):
    try:
        import tflite_runtime.interpreter as tflite
    except ModuleNotFoundError:
        from tensorflow import lite as tflite
else:
   from tensorflow import keras

[kahst]

Just pushed the conditional import, can you give it a go and see if it works? Seems to work fine on my machine, yet, I'm not using the TFLite runtime. I'll update the documentation accordingly.

[mcguirepr89]

The TFLite runtime is working great for me -- thanks!

[kahst]

I realized that TFLite-runtime 2.5.0 won't work, it has to be 2.7.0. Yet, this version doesn't seem to be available for all Python distros, e.g. the one I use on Windows. Therefore, I only recommend installing it on Ubuntu in the setup tutorial. Still closing this issue since it seems to work aside from that.

[fegue]

I'll test it as soon as I have a RPi near me.
I have only run the code with tflite-runtime 2.7.0 which is available on Raspbian Bullseye