Geek Answers Handbook

TensorFlow rebuild / deploy network without model?

I created and trained some networks with TensorFlow and successfully managed to save and restore model parameters.

However, for some scenarios - for example, deploying a trained network in a client infrastructure - this is not the best solution for delivering a complete code / model. So, I am wondering if there is a way to restore / run a trained network without the source code / model used for training?

I suppose this leads to the fact that TensorFlow can save a (compressed?) Version of the network architecture to checkpoint files in addition to variable weights.

How is this possible?

+4
source share
1

graphdef (*.pb), , , freeze_graph.py script : https://github.com/tensorflow/tensorflow/blob/master/tensorflow/python/tools/freeze_graph.py

script graphdef (.pb) (.ckpt) graphdef, ( script ).

+2

More articles:


All Articles