How to add "max_unpool2d" operator to onnxruntime?

[ArtyZe]

Hello,
Thanks for your great job. Now I have few questions about adding ops to onnxruntime.

1. As "max_unpool2d" is a torch op which is not able to run in onnxruntime, so I plan to add this op to onnxruntime. But I don't know how to write the source file and head file, especially compute function, as example of onnxruntime is useless now.
   
   Could you update it when you are convenient?
2. If my op domain is same as others, which should I use here?
   
   Thanks 👍

[jcwchen]

Hi @ArtyZe,

1. Sorry for the mistake. The right link is as follows:
   https://github.com/onnx/tutorials/blob/master/PyTorchCustomOperator/ort_custom_op/custom_op.cc

2. Will you use some custom opsets from their domain? If so, I think you should use the same op domain as theirs.

Thank you for reporting this document issue!

[ArtyZe]

Hi @ArtyZe,

1. Sorry for the mistake. The right link is as follows:
   https://github.com/onnx/tutorials/blob/master/PyTorchCustomOperator/ort_custom_op/custom_op.cc
2. Will you use some custom opsets from their domain? If so, I think you should use the same op domain as theirs.

Thank you for reporting this document issue!

Thanks for the help!
As what describled in doc,

now I compile it successfully. Got

So where should I put this lib file to support onnxruntime test model with max_unpool and other ops like conv2d? That means what should I do now to let my onnxruntime session call this function:

Actually cmake file link is also broken, so I have no idea compile my op to lib or bin file.

[jcwchen]

Hi @ArtyZe,
Sorry for the late reply. The right CmakeLists.txt link should be:
https://github.com/onnx/tutorials/blob/master/PyTorchCustomOperator/ort_custom_op/CMakeLists.txt
There are some urls broken as you mentioned. You can find the correct links in this PR: https://github.com/onnx/tutorials/pull/206/files

Hope it helps. Thanks.

[harshamedu]

Hi,

I am defining a model in onnx. I want to define a custom op. How can I do in python ?
I get 'error: PyOp is not a registered function/op' for the following code.

import onnx
import onnxruntime as ort

A = onnx.helper.make_tensor_value_info('A', onnx.TensorProto.FLOAT, [4])
B = onnx.helper.make_tensor_value_info('B', onnx.TensorProto.FLOAT, [4])
C = onnx.helper.make_tensor_value_info('C', onnx.TensorProto.FLOAT, [4])
D = onnx.helper.make_tensor_value_info('D', onnx.TensorProto.FLOAT, [4])
E = onnx.helper.make_tensor_value_info('E', onnx.TensorProto.FLOAT, [4])
F = onnx.helper.make_tensor_value_info('F', onnx.TensorProto.FLOAT, [4])

ad1_node = onnx.helper.make_node('Add', ['A', 'B'], ['S'])

mul_node = onnx.helper.make_node('Mul', ['C','D'], ['P'])

ad2_node = onnx.helper.make_node('Add', ['S', 'P'], ['H'])

py1_node = onnx.helper.make_node(op_type = 'PyOp', #required, must be 'PyOp'
inputs = ['H'], #required
outputs = ['F'], #required
domain = 'pyopadd_2', #required, must be unique
input_types = [onnx.TensorProto.FLOAT], #required
output_types = [onnx.TensorProto.FLOAT], #required
module = 'mymodule', #required
class_name = 'Add_2', #required
compute = 'compute') #optional, 'compute' by default

graph = onnx.helper.make_graph([ad1_node,mul_node,ad2_node, py1_node], 'multi_pyop_graph', [A,B,C,D], [F])
model = onnx.helper.make_model(graph, opset_imports=[onnx.helper.make_opsetid('pyopadd_2', 1)], producer_name = 'pyop_model')
onnx.save(model, './modeltemp.onnx')

ort_session = ort.InferenceSession('./modeltemp.onnx')
ort_output = ort_session.run(["F"], {'A':[1,2,3,4], 'B':[1,1,1,1], 'C':[2,2,2,2], 'D':[3,3,3,3]})
print(ort_output)

mymodule.py
class Add_2:
def compute(self, S):
return S+2

Thanks.