Support dynamic 'pads' and 'value' in Pad operator

[hariharans29]

Change description:

This change is to make two static attributes of the Pad op - 'pads' and 'value' into dynamic inputs to support conversion of some TensorFlow models into ONNX which allow these 2 values to be fed in dynamically to the op. This change is to basically support conversion of such TF models into ONNX.

Summary of changes:

• Make interface change in Pad operator as described in title
• Rename 'value' to 'constant_value' to reflect its usage only for 'constant' mode
• Make related node test case changes
• Make related shape inference test changes
• Make related changes in optimizer passes
• Make related optimizer test changes

Issues that are resolved:
Resolves #1765 - Adds description and examples for 'Edge' and 'Reflect' mode
Resolves #2313 - Allows testing optimizer(s) for previous opset specs of ops/nodes

[ebarsoum]

Resolve conflict, add example for other modes, and rename value to constant_value.

[Exlsunshine]

Hi @hariharans29, glad to see your change has been merged into master, but I still got the following error:

OP=Pad
Name=some/path/to/op/Pad
Inputs:
        some/path/to/op/concat:0=Concat, [-1, -1], 1
        some/path/to/op/Pad/paddings_Concat__113:0=Concat, [2, 2], 6
Outpus:
        some/path/to/op/Pad:0=[-1, -1], 1
Traceback (most recent call last):
  File "D:\Python\Python36\lib\site-packages\tf2onnx\tfonnx.py", line 352, in tensorflow_onnx_mapping
    func(g, node, **kwargs)
  File "D:\Python\Python36\lib\site-packages\tf2onnx\onnx_opset\nn.py", line 401, in version_1
    paddings = np.array(node.inputs[1].get_tensor_value()).transpose().flatten()
  File "D:\Python\Python36\lib\site-packages\tf2onnx\graph.py", line 256, in get_tensor_value
    raise ValueError("get tensor value: {} must be Const".format(self.name))
ValueError: get tensor value: some/path/to/op/Pad/paddings_Concat__113 must be Const
2019-11-21 15:52:39,788 - VERBOSE - tf2onnx.tfonnx: Mapping TF node to ONNX node(s)
2019-11-21 15:52:40,098 - ERROR - tf2onnx.tfonnx: Failed to convert node some/path/to/op/Fill


OP=ConstantOfShape
Name=some/path/to/op/Fill
Inputs:
        some/path/to/op/Fill__851:0=Cast, [1], 7
        some/path/to/op/strided_slice_21__842:0=Cast, [], 6
Outpus:
        some/path/to/op/Fill:0=[-1], 6
Traceback (most recent call last):
  File "D:\Python\Python36\lib\site-packages\tf2onnx\tfonnx.py", line 352, in tensorflow_onnx_mapping
    func(g, node, **kwargs)
  File "D:\Python\Python36\lib\site-packages\tf2onnx\onnx_opset\generator.py", line 100, in version_9
    value = np.array([node.inputs[1].get_tensor_value()]).astype(utils.map_onnx_to_numpy_type(dtype))
  File "D:\Python\Python36\lib\site-packages\tf2onnx\graph.py", line 256, in get_tensor_value
    raise ValueError("get tensor value: {} must be Const".format(self.name))
ValueError: get tensor value: some/path/to/op/strided_slice_21__842 must be Const

I am using:

python -m tf2onnx.convert --input a.pb --inputs in_0:0,in_1:0 --outputs o_0:0,o_1:0 --output b.onnx --opset 11 --fold_const  --continue_on_error --verbose

2019-11-21 15:52:31,785 - INFO - tf2onnx.tfonnx: Using tensorflow=1.13.1, onnx=1.6.0, tf2onnx=1.5.3/7b598d
2019-11-21 15:52:31,785 - INFO - tf2onnx.tfonnx: Using opset <onnx, 11>

Do you have any advice about this error? Thanks!