I work on micro controllers and teach JavaScript. I can convert a TFJS model to a tensorflow saved model and then to TFLITE and finally using XXD convert the binary file into a c-header file that can be used in an Arduino. Using the converter app this is very easy with these commands. Assuming the incoming TFJS model is called model.json
tensorflowjs_converter --input_format=tfjs_layers_model --output_format=keras_saved_model ./model.json ./
tflite_convert --keras_model_file ./ --output_file ./model.tflite
xxd -i model.tflite model.h
However, this entire process could be done automatically in Javascript in seconds, without loading the tensorflow_Converter app in some docker uploading the TFJS model, etc, etc.
For making and using machine learning models, being able to do the whole process with Javascript would be a huge assistance, and would simplify putting small ML models onto IOT internet of things) devices.
Why isn’t TensrflowJS_converter also written in Javascript? With some help I could probably do it for the basic models. Then I can tell people if my method does not work, try the official Tensoflow method. Can anyone point me in the right direction? Can it even be done?
Here is the github location of tfjs_converter, which is all in python. tfjs/tfjs-converter at master · tensorflow/tfjs · GitHub