Export Object Detector model to CoreML crashes

[apple]

Export Object Detector model to CoreML crashes

[helloniklas]

Further more, the model works all fine in Turi, I am able to make accurate predictions on new images. So not sure why this export to CoreML keeps crashing?

[gustavla]

@helloniklas Thanks for trying out the object detector! I'm sorry you are having this issue. I have never seen it before and it looks difficult for me to fix unless I can reproduce it. I suspect it may be something with your Python installation, so any additional information about your Python setup would help. For instance, is it the Python that comes with macOS or did you install it yourself (homebrew, macports, conda, etc.)?

Also, did you try export_coreml for any of the other models? For instance, does the image classifier work?

[helloniklas]

@gustavla So I haven't tried other models yet. I believe on this laptop python was installed with Homebrew, as well as the default Xcode install of python. However, I was under the impression that when following your install steps and setting it up with a virtualenv it would use the Python in that virtualenv. I'm new to the Python world. Is there a way I can ensure the virtualenv is using the correct python install?

Googling the error I get people do seem to mention it has to do with different python versions.

If I type which python it comes back with
/Users/niklas/development/turi/bin/python
And if I type python --version it comes back with
Python 2.7.14

[helloniklas]

@gustavla So after setting up the virtualenv with virtualenv --python=/usr/bin/python2.7 turiAgain
It now seems to work, CoreML model is saved out. Good times.

I guess without specifying the python path to use it used some incompatible version, although it still was python2.7

[srikris]

@helloniklas coremltools does not work with python installed via homebrew. That explains the error you are getting. I filed an issue with that repository (apple/coremltools#88)

[dmcgloin]

Got same error as @helloniklas. Everything worked in terms of training a model using the virtualenv (after lots of thrashing through docs - and me being unfamiliar with python).

But when it came to exporting to CoreML, I got the same fatal error.

Installing Anaconda resolved the issue for me.