Is it possible to convert keras model to coreml model if theano is used as backend?

I am trying to convert a `keras` model to `coreml`. It seems that my `theano` backend is creating a problem with the conversion. I have posted the problem here. I also tried upgrading `theano`. I read here that I might require to change the ordering of data as `(height, width, channels)`. However my problem seems to be that `cormltools` is not able to identify my backend. Does this conversion work with `theano` backend or does it require `tensorflow`?

Accepted Reply

Currently, the Keras conversion of coremltools requires you to have the tensorflow backend. We are working hard to get theano backend working too. Stay tuned for updates.

Replies

Currently, the Keras conversion of coremltools requires you to have the tensorflow backend. We are working hard to get theano backend working too. Stay tuned for updates.

Thank You for the reply.

I know that `keras 1.2.2` is required to use `coremltools`. Is it necessary to use any particular version of `tensorflow` and `python`?

Tensorflow version 1.0.x and 1.1.x are supported and the latest version of coremltools (0.4.0) supports Keras 2.0.0+. Currently only python 2.7 is supported but we are working on support for Python 3 and plan to have it in an upcoming update

Okay, I'm confused.... What is supported, as of today; July 31, 2017? I'm running keras 2.0.6 with Theano backend, but I can't seem to convert to coreml (AttributeError: 'module' object has no attribute 'convert'). I'm guessing this is because of the Theano backend?

just lending my voice to the crowd that really wants theano to be supported by coremltools