No schema registered for 'Celu'

[bratyslav]

Describe the bug

Error "No schema registered for 'Celu'!" happens when I'm trying to run "onnx_tf.backend.prepare(my_model)". Is there a way to solve this problem?

To Reproduce

My code is just: onnx_tf.backend.prepare(onnx.load(PATH))

ONNX model file

https://drive.google.com/file/d/1CkXFBeEjLFqmdDcPYb77Wqz_rPYrpdOw/view?usp=sharing

Python, ONNX, ONNX-TF, Tensorflow version

• Python version: 3.6.9
• ONNX version: 1.8.1
• ONNX-TF version: 1.7.0
• Tensorflow version: 2.4.1

Additional context

[chinhuang007]

Please specify Tensorflow version. Use 2.4.x and see if it helps.

[bratyslav]

Please specify Tensorflow version. Use 2.4.x and see if it helps.

Tensorflow version is 2.4.1. (I edited the issue).

[chinhuang007]

I tried with the same env and the prepare call works. However, the export_graph call fails with "RuntimeError: Resize coordinate_transformation_mode=pytorch_half_pixel is not supported in Tensorflow"

Unfortunately, that is a known limitation in resize conversion, due to no Tensorflow equivalent for the pytorch_half_pixel option.

[bratyslav]

Okey, I will try to fix it, thank you for attention.

[mavaras]

Hi. I'm having the exact same issue as described here. With my .onnx model, when I call prepare I get "Error "No schema registered for 'Celu'!" I have the same exact libraries versions as described above onnx model: https://drive.google.com/file/d/1ukBrN3zGAkI368PFjtZ4Olx2RCyc8rGh/view?usp=sharing

Can anyone help? Thanks in advance

[mavaras]

Also, for @bratyslav model I get the same error. Just doing

onnx_model = onnx.load(f'{root_dir}/models/modnet.onnx')
tf_rep = prepare(onnx_model)

[palsol]

I am porting model from pytorch to tflite. I had same problem with Celu, i changed opset_version in torch.onnx.export (from 11 to 12) and celu error gone.

Now i having another error mentioned before (RuntimeError: Resize coordinate_transformation_mode=pytorch_half_pixel is not supported in Tensorflow.).

[chinhuang007]

As you can tell here, https://github.com/onnx/onnx/blob/master/docs/Operators.md#Celu, the Celu op is first introduced in ONNX opset 12. The initial problem is in the Pytorch exporter, which could produce a model with opset < 12 and including Celu. The default opset number is way too old as seen in this issue, https://github.com/pytorch/pytorch/issues/48341

[Jeffwhen]

I encountered same problem. But my model does not have any Celu op, so I am very confused. Here is stack dump

Invalid MIT-MAGIC-COOKIE-1 key2021-03-19 17:47:09,586 - onnx-tf - INFO - Start converting onnx pb to tf pb:
Traceback (most recent call last):
  File "/home/jeffwhen/.local/bin/onnx-tf", line 8, in <module>
    sys.exit(main())
  File "/home/jeffwhen/.local/lib/python3.8/site-packages/onnx_tf/cli.py", line 20, in main
    return onnx_tf.converter.main(args[1:])
  File "/home/jeffwhen/.local/lib/python3.8/site-packages/onnx_tf/converter.py", line 20, in main
    convert(**{k: v for k, v in vars(args).items() if v is not None})
  File "/home/jeffwhen/.local/lib/python3.8/site-packages/onnx_tf/converter.py", line 134, in convert
    tf_rep = backend.prepare(onnx_model, **kwargs)
  File "/home/jeffwhen/.local/lib/python3.8/site-packages/onnx_tf/backend.py", line 70, in prepare
    return cls.onnx_model_to_tensorflow_rep(model, strict, **kwargs)
  File "/home/jeffwhen/.local/lib/python3.8/site-packages/onnx_tf/backend.py", line 90, in onnx_model_to_tensorflow_rep
    return cls._onnx_graph_to_tensorflow_rep(model.graph, opset_import, strict,
  File "/home/jeffwhen/.local/lib/python3.8/site-packages/onnx_tf/backend.py", line 111, in _onnx_graph_to_tensorflow_rep
    handlers = cls._get_handlers(opset)
  File "/home/jeffwhen/.local/lib/python3.8/site-packages/onnx_tf/backend.py", line 303, in _get_handlers
    return get_all_backend_handlers(opset_dict)
  File "/home/jeffwhen/.local/lib/python3.8/site-packages/onnx_tf/common/handler_helper.py", line 25, in get_all_backend_handlers
    since_version = defs.get_schema(
onnx.onnx_cpp2py_export.defs.SchemaError: No schema registered for 'Celu'!

[chinhuang007]

@Jeffwhen please provide your onnx file for debugging. Also your environment details, such as onnx, tensorflow versions, thanks.

[Jeffwhen]

@chinhuang007 onnx version 1.8.1 tensorflow 2.3.1 onnx-tf 1.7.0. Model file on google drive

[chinhuang007]

@Jeffwhen The issue is in the combination of onnx 1.8.1 and onnx-tf 1.7.0. The onnx-tf 1.7.0 from pip install officially supports onnx 1.7.0. You have two options to get rid of this error. 1. install onnx 1.7.0 2. source install onnx-tf from the master branch.

Once you have done that, unfortunately you will hit a known unsupported case "Resize coordinate_transformation_mode=pytorch_half_pixel is not supported in Tensorflow". The only solution to that is to use one of the supported resize options, seen in the notes section of https://github.com/onnx/onnx-tensorflow/blob/master/doc/support_status_v1_7_0.md

[Adhithya-tech]

Okey, I will try to fix it, thank you for attention.

Could you tell me how you did? @bratyslav