I'm currently looking to use Microsoft Cognitive Service custom vision. It comes with a prediction API which you can use to fetch the predicted results. Is it possible to use Custom Vision without using this prediction API? With that, I mean just export the model to Tensorflow and build my own API which fetches the results directly from Tensorflow.
It seems like I can export the model for mobile usage or to run it in a Docker container. However it seems like the prediction API is wired into that Docker container.