fumihwh
commented
5 years ago @destindof
The original graph doesn't have shape information.
pb_wrapper.py L139. Replace _add_infer_shapes with following.
@staticmethod
def _add_infer_shapes(graph_def, input_shape=[1, 224, 224, 3]):
with tf.Graph().as_default():
with tf.Session(
config=tf.ConfigProto(
graph_options=tf.GraphOptions(infer_shapes=True))) as sess:
if input_shape:
input_node = [
node for node in graph_def.node if node.name == "input"
][0]
for i in input_shape:
dim = TensorShapeProto.Dim()
dim.size = i
input_node.attr["shape"].shape.dim.extend([dim])
tf.import_graph_def(graph_def, name="")
return sess.graph_def
destindof
XUYUNYUN666
Describe the bug
When I try to convert the TensorFlow in pb format to ONNX model using the command onnx-tf convert -t onnx -i mobilenet_v2_1.4_224_frozen.pb -o output.onnx
The error shows up: INFO:root:Start converting tf pb to onnx pb: 2019-03-06 16:35:48.043052: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 FMA Traceback (most recent call last): File "/home/ycxu/.local/bin/onnx-tf", line 11, in
load_entry_point('onnx-tf', 'console_scripts', 'onnx-tf')()
File "/home/ycxu/onnx-tensorflow-master/onnx_tf/cli.py", line 22, in main
return onnx_tf.converter.main(args[1:])
File "/home/ycxu/onnx-tensorflow-master/onnx_tf/converter.py", line 24, in main
convert({k: v for k, v in vars(args).items() if v is not None})
File "/home/ycxu/onnx-tensorflow-master/onnx_tf/converter.py", line 249, in convert
onnx_model = frontend.tensorflow_graph_to_onnx_model(graph_def, kwargs)
File "/home/ycxu/onnx-tensorflow-master/onnx_tf/frontend.py", line 145, in tensorflow_graph_to_onnx_model
ignore_unimplemented, optimizer_passes)
File "/home/ycxu/onnx-tensorflow-master/onnx_tf/frontend.py", line 158, in _make_onnx_model
ignore_unimplemented)
File "/home/ycxu/onnx-tensorflow-master/onnx_tf/frontend.py", line 93, in tensorflow_graph_to_onnx_graph
data_type_cast_map=onnx_graph.data_type_cast_map)
File "/home/ycxu/onnx-tensorflow-master/onnx_tf/handlers/frontend_handler.py", line 41, in handle
return super(FrontendHandler, cls).handle(node, kwargs)
File "/home/ycxu/onnx-tensorflow-master/onnx_tf/handlers/handler.py", line 60, in handle
return ver_handle(node, kwargs)
File "/home/ycxu/onnx-tensorflow-master/onnx_tf/handlers/frontend/convolution.py", line 32, in version_1
return cls.conv_op(node, d=d, **kwargs)
File "/home/ycxu/onnx-tensorflow-master/onnx_tf/handlers/frontend/conv_mixin.py", line 30, in conv_op
spatial_indices))
File "/home/ycxu/onnx-tensorflow-master/onnx_tf/handlers/frontend/conv_mixin.py", line 29, in
map(lambda i: node_dict[node.inputs[0]].attr["_output_shapes"][0][i],
IndexError: list index out of range
ONNX model file Downloaded at https://storage.googleapis.com/mobilenet_v2/checkpoints/mobilenet_v2_1.4_224.tgz
Python, ONNX, ONNX-TF, Tensorflow version
Python 3.6.7 (default, Oct 22 2018, 11:32:17) [GCC 8.2.0]