traceback (after attempting to convert custom MobileNet model)
Traceback (most recent call last):
File "coremlconverte2r.py", line 42, in <module>
class_labels=class_labels)
File "/Users/user147784/Library/Python/2.7/lib/python/site-packages/coremltools/converters/keras/_keras_converter.py", line 752, in convert
custom_conversion_functions=custom_conversion_functions)
File "/Users/user147784/Library/Python/2.7/lib/python/site-packages/coremltools/converters/keras/_keras_converter.py", line 534, in convertToSpec
from . import _keras2_converter
File "/Users/user147784/Library/Python/2.7/lib/python/site-packages/coremltools/converters/keras/_keras2_converter.py", line 14, in <module>
from . import _layers2
File "/Users/user147784/Library/Python/2.7/lib/python/site-packages/coremltools/converters/keras/_layers2.py", line 10, in <module>
from keras_applications.mobilenet import relu6
ImportError: cannot import name relu6
Tried the following solution
from keras.utils.generic_utils import CustomObjectScope
with CustomObjectScope({'relu6': keras.applications.mobilenet.relu6,'DepthwiseConv2D': keras.applications.mobilenet.DepthwiseConv2D}):
Still received error
WARNING:root:Keras version 2.2.4 detected. Last version known to be fully compatible of Keras is 2.1.6 .
WARNING:root:TensorFlow version 1.11.0 detected. Last version known to be fully compatible is 1.5.0 .
Traceback (most recent call last):
File "coremlconverte2r.py", line 12, in <module>
with CustomObjectScope({'relu6': keras.applications.mobilenet.relu6,'DepthwiseConv2D': keras.applications.mobilenet.DepthwiseConv2D}):
AttributeError: 'module' object has no attribute 'relu6'
Have not even the slightest of idea whats causing this.
Whats the version of Keras installed on your system?
Can you build from top of master and try again?
This error should be resolved after PR #278 went in.
Latest version of keras (2.2.4)
Currently on a virtual mac environment so it has python 2.7.10 and 3.7
Trying to build from master, not familiar with Cmake
Would correct statement be:
cmake . DPYTHON=$(python)
Just execute these commands:
git clone https://github.com/apple/coremltools.git
cd coremltools
cmake . -DPYTHON=$(which python) -DPYTHON_CONFIG=$(which python-config)
make
pip install -e .
Thanks for the help.
for anyone else this will do the trick also
pip install -U https://github.com/apple/coremltools/archive/master.zip --user
@zantism Using pip install would omit the native libraries used by Core ML. Recompiling from source is a better option.
Also take note of #295 (missing MLFeatureTypeSequence symbol) when building from source.