When to convert tensorflow mobilenet frozen graph, occurred error

[FrozenGene]

The mobilenet model: https://storage.googleapis.com/download.tensorflow.org/models/mobilenet_v1_1.0_224_frozen.tgz

when to convert it, prompt:
File "/usr/local/lib/python3.6/site-packages/tf2onnx-0.0.0.1-py3.6.egg/tf2onnx/convert.py", line 67, in
File "/usr/local/lib/python3.6/site-packages/tf2onnx-0.0.0.1-py3.6.egg/tf2onnx/convert.py", line 56, in main
File "/usr/local/lib/python3.6/site-packages/tf2onnx-0.0.0.1-py3.6.egg/tf2onnx/tfonnx.py", line 1000, in process_tf_graph
File "/usr/local/lib/python3.6/site-packages/tf2onnx-0.0.0.1-py3.6.egg/tf2onnx/tfonnx.py", line 939, in tensorflow_onnx_mapping
File "/usr/local/lib/python3.6/site-packages/tf2onnx-0.0.0.1-py3.6.egg/tf2onnx/tfonnx.py", line 521, in relu6_op
ValueError: negative dimensions are not allowed

[guschmue]

I changed the default for unknown dimension from -1 to 1 in case the runtime does not support -1 (onnx-caffe2 ie. does not support -1)

[leeqiaogithub]

I have the same problem when convert mobilenet_v2_1.0_224.pb to onnx.how can i do to resolve it?

[guschmue]

different issue related to relu6.
Let me see if I can fix it.

[guschmue]

@leeqiaogithub , I just made some change and the model at least will convert. You could give it a try.
My runtime barks at it, need to look at why but you could try it on your runtime.