TypeError: must be str, not NoneType

[nrauch]

I am trying to convert a Keras model developed with Tensorflow 2.3.1. When converting the model using onnx_model = keras2onnx.convert_keras(model, 'fasttext-onnx', debug_mode=1) I get the error in the title. Following the entire error track:

TypeError                                 Traceback (most recent call last)
<ipython-input-14-2989974affce> in <module>
      1 output_model_path = "fasttext.onnx"
----> 2 onnx_model = keras2onnx.convert_keras(model, 'fasttext-onnx', debug_mode=1)

/opt/conda/lib/python3.7/site-packages/keras2onnx/main.py in convert_keras(model, name, doc_string, target_opset, channel_first_inputs, debug_mode, custom_op_conversions)
     78     input_names = []
     79     output_names = []
---> 80     output_dict = {}
     81     if is_tf2 and is_tf_keras:
     82         tf_graph = build_layer_output_from_model(model, output_dict, input_names,

/opt/conda/lib/python3.7/site-packages/keras2onnx/parser.py in parse_graph(topo, graph, target_opset, output_names, keras_node_dict)
    839 def parse_graph(topo, graph, target_opset, output_names, keras_node_dict):
    840     # type: (Topology, tf.Graph, int, [], []) -> Topology
--> 841     """
    842     Build the node-layer mapper and parse the whole TF graph of Keras Model.
    843     """

/opt/conda/lib/python3.7/site-packages/keras2onnx/parser.py in _parse_graph_core_v2(graph, keras_node_dict, topology, top_scope, output_names)
    716             layer_info = LayerInfo.create(graph, node, layer_key, keras_node_dict,
    717                                           current_layer_inputs, current_layer_outputs, inference_nodeset)
--> 718 
    719     nodelist = []
    720     layer_inputs = _visit_nodelist(layer_info.nodelist, graph_inputs, None, keras_node_dict, node, nodelist,

/opt/conda/lib/python3.7/site-packages/keras2onnx/parser.py in _parse_nodes_v2(graph, inference_nodeset, graph_inputs, keras_node_dict, node, varset, visited, q_overall)
    666         output_mask_list = kenode.outbound_layer.output_mask if \
    667             isinstance(kenode.outbound_layer.output_mask, list) else [kenode.outbound_layer.output_mask]
--> 668         for ts_ in output_mask_list:
    669             if ts_ is not None:
    670                 mask_ts.append(ts_)

/opt/conda/lib/python3.7/site-packages/keras2onnx/_parser_tf.py in create(node, layer, outputs_map, inference_nodeset)
    128         while next_itr:
    129             visited |= next_itr
--> 130             next_itr.clear()
    131             for n_ in visited:
    132                 for i_ in n_.inputs:

TypeError: must be str, not NoneType

It is not clear to me what it is None and I hope you can help me.

Thanks!

[tolstikh]

Same issue here

[dnola]

Also experiencing this same issue! A Keras.applications ResNet50 model that is first saved to disk and reloaded seems to break keras2onnx.

Any help would be appreciated. The same model converts properly in tf2onnx.

[Spritaro]

I had a similar problem. I was trying to copy weights from model1 to model2 and convert the model2 to ONNX, then it raised the error must be str, not NoneType. Strangely, I was able to avoid the error by creating model2 before model1.