Unsupported shape calculation for operator BatchNormalization

I made a new model using keras, and saved it to an .hdf file using a callback during training.

I reloaded the model and tried to convert to an ONNX model:

model = keras.models.load_model(filename)
convert_model = winmltools.convert_keras(keras_model, 7)
winmltools.save_model(convert_model, onnx_model_name)

convert_keras throws an error here. Any idea what I might be doing wrong? python 3.6.2, Tensorflow 1.8, Keras 2.1.5, onnxmltools 1.3.2

The keras-retinanet implementation uses keras_resnet which has BatchNormalization referred to in the error.

Error below:

ValueError                                Traceback (most recent call last)
<ipython-input-11-800754ba9409> in <module>()
----> 1 convert_to_onnx(model, "snapshots/onnxmodel.onnx")

<ipython-input-9-d58ccba40905> in convert_to_onnx(keras_model, onnx_model_name)
      2 
      3         ########  Conver to ONNX ############
----> 4         convert_model = winmltools.convert_keras(keras_model, 7)
      5         winmltools.save_model(convert_model, onnx_model_name)

~/.local/lib/python3.6/site-packages/winmltools/convert/main.py in convert_keras(model, target_opset, name, initial_types, doc_string, default_batch_size, channel_first_inputs, custom_conversion_functions, custom_shape_calculators)
    151     return _convert_keras(model, name=name, target_opset=target_opset, initial_types=initial_types, doc_string=doc_string,
    152                           default_batch_size=default_batch_size, channel_first_inputs=channel_first_inputs,
--> 153                           custom_conversion_functions=custom_conversion_functions, custom_shape_calculators=custom_shape_calculators)
    154 
    155 

~/.local/lib/python3.6/site-packages/onnxmltools/convert/keras/convert.py in convert(model, name, default_batch_size, initial_types, doc_string, target_opset, targeted_onnx, channel_first_inputs, custom_conversion_functions, custom_shape_calculators)
     41                            custom_conversion_functions, custom_shape_calculators)
     42 
---> 43     topology.compile()
     44 
     45     if name is None:

~/.local/lib/python3.6/site-packages/onnxmltools/convert/common/_topology.py in compile(self)
    623         self._resolve_duplicates()
    624         self._fix_shapes()
--> 625         self._infer_all_types()
    626         self._check_structure()
    627 

~/.local/lib/python3.6/site-packages/onnxmltools/convert/common/_topology.py in _infer_all_types(self)
    499                 pass  # in Keras converter, the shape calculator can be optional.
    500             else:
--> 501                 operator.infer_types()
    502 
    503     def _resolve_duplicates(self):

~/.local/lib/python3.6/site-packages/onnxmltools/convert/common/_topology.py in infer_types(self)
    101     def infer_types(self):
    102         # Invoke a core inference function
--> 103         _registration.get_shape_calculator(self.type)(self)
    104 
    105 

~/.local/lib/python3.6/site-packages/onnxmltools/convert/common/_registration.py in get_shape_calculator(operator_name)
     66     '''
     67     if operator_name not in _shape_calculator_pool:
---> 68         raise ValueError('Unsupported shape calculation for operator %s' % operator_name)
     69     return _shape_calculator_pool[operator_name]

ValueError: Unsupported shape calculation for operator <class 'keras_resnet.layers._batch_normalization.BatchNormalization'>

[jiafatom]

The keras2onnx converter calls tf2onnx converter to process some layers. This error throws from tf2onnx and indicates that the graph has cycles (so it cannot handle). Does this graph contain cycles?

[hotamy]

@scutzhe I have same problem.Is this problem solved?