None values not supported when converting onnx to tensorflow

[padmasreenagarajan]

Im trying to convert Mobilenetv3 onnx model to tensorflow.
I need tensorflow==1.12.0 for my inferencing.
I searched all over to find the onnx and onnx-tf versions corresponding to tensorflow==1.12.0

Could anyone please suggest me the compatible versions of onnx,onnx-tf and torch for my required tensorflow version?

[pluradj]

Did you test converting your model with onnx 1.3.0 and onnx-tensorflow 1.3.0 as I mentioned previously? If it failed, could you provide more details on what error you were seeing?

[padmasreenagarajan]

@pluradj
Initially i faced 2 errors-
1)AttributeError: module 'tensorflow._api.v1.math' has no attribute 'is_inf'
which is solved by modifying the code in onnx_tf/handlers/backend/is_inf.py

2)onnx.onnx_cpp2py_export.checker.ValidationError: base_net.0.bias in initializer but not in graph input
which is solved by exporting to onnx with Keep_initializers_as _inputs==True

And finally,i am unable to solve this error
onnx.onnx_cpp2py_export.checker.ValidationError: Your model ir_version is higher than the checker's.

I used
Torch==1.3.0+cpu;Torchvision==0.4.0+cpu
onnx==1.3.0
onnx-tf==1.3.0
tensorflow==1.12.0

Do guide me in solving this!@pluradj

[pluradj]

What is the ir_version of your model? You'll need to make sure you're targeting a compatible version of ONNX.

[pluradj]

For reference:

ONNX version	IR version
v1.0 - v1.3.0	3
v1.4.0	4
v1.5.0	5
v1.6.0	6
v1.7.0	7

[padmasreenagarajan]

my model's ir version is 3
@pluradj

[pluradj]

Hmm. ONNX 1.3.0 supports IR version 3, so you shouldn't be getting the error "Your model ir_version is higher than the checker's." Could you please share your ONNX model?

[padmasreenagarajan]

Hi..Im facing this error when converting Mobilenetv3 onnx file to tensorflow

File "Python36\site-packages\tensorflow_core\python\framework\tensor_util.py", line 437, in make_tensor_proto
raise ValueError("None values not supported.")

ValueError: None values not supported.
Do guide me in solving this..
@pluradj
Im using
torch==1.3.0+cpu
onnx==1.5.0
onnx-tf==1.5.0
tensorflow==1.15.0
opset==10

[pluradj]

Could you please share your ONNX model?

[padmasreenagarajan]

Sorry @pluradj
As it is confidential, I am unable to share it here

[pluradj]

could you at least share a more complete stack trace?

[padmasreenagarajan]

This is the entire error and warning messages.
As I am new to onnx and tensorflow,
Could u please guide me in resolving this? @pluradj

C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\common\handler_helper.py:37: UserWarning: Unknown op ConstantFill in domain ai.onnx.
handler.ONNX_OP, handler.DOMAIN or "ai.onnx"))
C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\common\handler_helper.py:37: UserWarning: Unknown op GatherND in domain ai.onnx.
handler.ONNX_OP, handler.DOMAIN or "ai.onnx"))
C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\common\handler_helper.py:37: UserWarning: Unknown op ImageScaler in domain ai.onnx.
handler.ONNX_OP, handler.DOMAIN or "ai.onnx"))
C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\common\handler_helper.py:37: UserWarning: Unknown op Range in domain ai.onnx.
handler.ONNX_OP, handler.DOMAIN or "ai.onnx"))
C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\common\handler_helper.py:37: UserWarning: Unknown op Round in domain ai.onnx.
handler.ONNX_OP, handler.DOMAIN or "ai.onnx"))
C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\common\handler_helper.py:37: UserWarning: Unknown op ScatterElements in domain ai.onnx.
handler.ONNX_OP, handler.DOMAIN or "ai.onnx"))
C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\common\handler_helper.py:37: UserWarning: Unknown op ScatterND in domain ai.onnx.
handler.ONNX_OP, handler.DOMAIN or "ai.onnx"))
WARNING:tensorflow:From C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\backend.py:123: The name tf.placeholder is deprecated. Please use tf.compat.v1.placeholder instead.

2020-09-23 11:39:39.543468: I tensorflow/core/platform/cpu_feature_guard.cc:142] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2
2020-09-23 11:39:39.616856: W tensorflow/stream_executor/platform/default/dso_loader.cc:55] Could not load dynamic library 'nvcuda.dll'; dlerror: nvcuda.dll not found
2020-09-23 11:39:39.617362: E tensorflow/stream_executor/cuda/cuda_driver.cc:318] failed call to cuInit: UNKNOWN ERROR (303)
2020-09-23 11:39:39.650977: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:169] retrieving CUDA diagnostic information for host: 6011052LD01
2020-09-23 11:39:39.651319: I tensorflow/stream_executor/cuda/cuda_diagnostics.cc:176] hostname: 6011052LD01
WARNING:tensorflow:From C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\handlers\backend_handler.py:182: div (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Deprecated in favor of operator or tf.math.divide.
WARNING:tensorflow:From C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\handlers\backend\reshape.py:26: where (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.where in 2.0, which has the same broadcast rule as np.where
WARNING:tensorflow:From C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\handlers\backend\reshape.py:31: sparse_to_dense (from tensorflow.python.ops.sparse_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Create a tf.sparse.SparseTensor and use tf.sparse.to_dense instead.
WARNING:tensorflow:From C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\handlers\backend\gemm.py:14: flatten (from tensorflow.python.layers.core) is deprecated and will be removed in a future version.
Instructions for updating:
Use keras.layers.flatten instead.
WARNING:tensorflow:From C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\tensorflow_core\python\layers\core.py:332: Layer.apply (from tensorflow.python.keras.engine.base_layer) is deprecated and will be removed in a future version.
Instructions for updating:
Please use layer.__call__ method instead.
Traceback (most recent call last):
File "C:/Users/PycharmProjects/mbv3/Scripts/convert_to_onnx.py", line 252, in
tf_rep = prepare(onnx_model) # prepare tf representation
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\backend.py", line 65, in prepare
return cls.onnx_model_to_tensorflow_rep(model, strict)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\backend.py", line 85, in onnx_model_to_tensorflow_rep
return cls._onnx_graph_to_tensorflow_rep(model.graph, opset_import, strict)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\backend.py", line 143, in _onnx_graph_to_tensorflow_rep
onnx_node, tensor_dict, handlers, opset=opset, strict=strict)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\backend.py", line 238, in _onnx_node_to_tensorflow_op
return handler.handle(node, tensor_dict=tensor_dict, strict=strict)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\handlers\handler.py", line 61, in handle
return ver_handle(node, **kwargs)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx_tf\handlers\backend\slice.py", line 56, in version_10
input_tensor_shape = tf.constant(input_tensor.shape.dims, ends.dtype)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\tensorflow_core\python\framework\constant_op.py", line 161, in constant_v1
allow_broadcast=False)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\tensorflow_core\python\framework\constant_op.py", line 265, in _constant_impl
allow_broadcast=allow_broadcast))
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\tensorflow_core\python\framework\tensor_util.py", line 437, in make_tensor_proto
raise ValueError("None values not supported.")

ValueError: None values not supported.

[pluradj]

The last error looks like a problem with the Slice operator and dynamic size support. A fix for that was merged into tf-1.x branch with PR #566. Try upgrading your packages to ONNX 1.6.0 and ONNX-TF 1.6.0 to pick up this fix. The git tag for that release is v1.6.0-tf-1.15. ONNX-TF 1.6.0 isn't available on PyPI currently, but you can install it like this:

pip install git+https://github.com/onnx/onnx-tensorflow.git@v1.6.0-tf-1.15

[padmasreenagarajan]

Hi @pluradj ,
I have already tried the versions mentioned onnx==onnx-tf==1.6.0 with tf=1.15.
However I face the following error for opset 11

Traceback (most recent call last):
File "C:/Users/PycharmProjects/mbv3/Scripts/convert_to_onnx.py", line 249, in
torch.onnx.export(net, example, 'C:/Users/PycharmProjects/mbv3/mb3-ssd-lite-Epoch-170-Loss-inf-9-tf1.15-onnx1.6-onnx_tf1.6-torch1.3.onnx', opset_version=11)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\onnx_init_.py", line 143, in export
strip_doc_string, dynamic_axes, keep_initializers_as_inputs)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\onnx\utils.py", line 66, in export
dynamic_axes=dynamic_axes, keep_initializers_as_inputs=keep_initializers_as_inputs)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\onnx\utils.py", line 382, in _export
fixed_batch_size=fixed_batch_size)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\onnx\utils.py", line 262, in _model_to_graph
fixed_batch_size=fixed_batch_size)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\onnx\utils.py", line 132, in _optimize_graph
graph = torch._C.jit_pass_onnx(graph, operator_export_type)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\onnx_init.py", line 174, in _run_symbolic_function
return utils._run_symbolic_function(*args, **kwargs)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\onnx\utils.py", line 619, in _run_symbolic_function
return op_fn(g, *inputs, **attrs)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\torch\onnx\symbolic_helper.py", line 175, in symbolic_fn
.format(name, _export_onnx_opset_version))
RuntimeError: ONNX export failed on hardtanh, which is not implemented for opset 11. Try exporting with other opset versions.

and the following for opset 9

Traceback (most recent call last):
File "C:/Users/PycharmProjects/mbv3/Scripts/convert_to_onnx.py", line 251, in
checker.check_graph(onnx_model.graph)
File "C:\Users\AppData\Local\Programs\Python\Python36\lib\site-packages\onnx\checker.py", line 53, in checker
proto.SerializeToString(), ctx)
onnx.onnx_cpp2py_export.checker.ValidationError: Unrecognized attribute: max for operator Clip

==> Context: Bad node spec: input: "482" output: "483" op_type: "Clip" attribute { name: "max" f: 6 type: FLOAT } attribute { name: "min" f: 0 type: FLOAT }

Let me know if u need the complete stack trace.

BTW, tensorflow (.pb) file got generated successfully with
onnx=1.5.0
onnx-tf==1.6.0 (from tf-1.x branch)
tf==1.15.0
torch==1.3.0
opset==10
But this does not follow the version matrix. Please do let me know how it got generated successfully despite not following the version matrix
Thank you!