tflite model maker tools
Tensorflow made the tflite model maker available a while go, which (according to the project website) simplifies the process of training a TensorFlow Lite model using custom dataset. It uses transfer learning to reduce the amount of training data required and shorten the training time.
But instead of falling back on scripting Python again, we pulled code from the their example notebooks into a library that offers command-line utilies and published it on PyPI to avoid having to write Python code over and over again for every new problem/dataset. It also makes it easy to create docker images from it, making it highly reusable.