How to use code with onnx?

[za13]

I'm using the code in the notebook in the "examples"

I'm trying to use keras2onnx with it also. I tried

import keras2onnx
import onnxruntime

# convert to onnx model
onnx_model = keras2onnx.convert_keras(model, model.name)

# runtime prediction
content = onnx_model.SerializeToString()
sess = onnxruntime.InferenceSession(content)
x = x if isinstance(x, list) else [x]
feed = dict([(input.name, x[n]) for n, input in enumerate(sess.get_inputs())])
pred_onnx = sess.run(None, feed)

But I got the following error:

InvalidArgumentError                      Traceback (most recent call last)
~/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/importer.py in import_graph_def(graph_def, input_map, return_elements, name, op_dict, producer_op_list)
    426         results = c_api.TF_GraphImportGraphDefWithResults(
--> 427             graph._c_graph, serialized, options)  # pylint: disable=protected-access
    428         results = c_api_util.ScopedTFImportGraphDefResults(results)

InvalidArgumentError: Node 'block1b_drop/cond/mul/y': Unknown input node '^block1b_drop/cond/switch_t'

During handling of the above exception, another exception occurred:

ValueError                                Traceback (most recent call last)
<ipython-input-16-a5d3ba8b443c> in <module>
      3 
      4 # convert to onnx model
----> 5 onnx_model = keras2onnx.convert_keras(model, model.name)
      6 
      7 # runtime prediction

~/anaconda3/lib/python3.6/site-packages/keras2onnx/main.py in convert_keras(model, name, doc_string, target_opset, channel_first_inputs, debug_mode, custom_op_conversions)
     98                         custom_op_dict=custom_op_conversions)
     99     topology.debug_mode = debug_mode
--> 100     parse_graph(topology, sess.graph, target_opset, output_names)
    101     topology.compile()
    102 

~/anaconda3/lib/python3.6/site-packages/keras2onnx/parser.py in parse_graph(topo, graph, target_opset, output_names)
    647         topo.raw_model.add_input_name(str_value)
    648 
--> 649     return _parse_graph_scope(graph, keras_layer_ts_map, topo, top_level, output_names)

~/anaconda3/lib/python3.6/site-packages/keras2onnx/parser.py in _parse_graph_scope(graph, keras_node_dict, topology, top_scope, output_names)
    597             _convert_keras_timedistributed(graph, nodes, layer_key_, model_, varset)
    598         elif layer_key_ is None or get_converter(type(layer_key_)) is None:
--> 599             _convert_general_scope(nodes, varset)
    600         else:
    601             _convert_keras_scope(graph, nodes, layer_key_, model_, varset)

~/anaconda3/lib/python3.6/site-packages/keras2onnx/parser.py in _convert_general_scope(node_list, varset)
    299 
    300     sess = keras.backend.get_session()
--> 301     subgraph, replacement = create_subgraph(sess.graph, node_list, sess, operator.full_name)
    302     setattr(operator, 'subgraph', subgraph)
    303     vars_, ts = _locate_inputs_by_node(node_list, varset)

~/anaconda3/lib/python3.6/site-packages/keras2onnx/subgraph.py in create_subgraph(tf_graph, node_list, sess, dst_scope)
    135     with tf.Graph().as_default() as sub_graph:
    136         im_scope = "" if dst_scope is None else dst_scope
--> 137         tf.import_graph_def(output_graph_def, name=im_scope)
    138         if im_scope:
    139             replacement = {k_: im_scope + '/' + k_ for k_ in replacement}

~/anaconda3/lib/python3.6/site-packages/tensorflow/python/util/deprecation.py in new_func(*args, **kwargs)
    505                 'in a future version' if date is None else ('after %s' % date),
    506                 instructions)
--> 507       return func(*args, **kwargs)
    508 
    509     doc = _add_deprecated_arg_notice_to_docstring(

~/anaconda3/lib/python3.6/site-packages/tensorflow/python/framework/importer.py in import_graph_def(graph_def, input_map, return_elements, name, op_dict, producer_op_list)
    429       except errors.InvalidArgumentError as e:
    430         # Convert to ValueError for backwards compatibility.
--> 431         raise ValueError(str(e))
    432 
    433     # Create _DefinedFunctions for any imported functions.

ValueError: Node 'block1b_drop/cond/mul/y': Unknown input node '^block1b_drop/cond/switch_t'

[JordanMakesMaps]

I haven't used Onyx in a while but it seems like they've progressed a lot since before. Previously I tried to convert a custom ResNet50 from Pytorch to Keras and the associated weights. Because the model had a few custom layers, Onyx was not capable of converting the architecture because it didn't know how to convert those layers.

What @qubvel has here in this repo might not be convertible with Onyx just yet, unless you're willing to write some code that will handle it. Otherwise I'd recommend looking at some of architectures that have been shown to work in Onyx format in the model zoo.