NotImplementedError: BatchNormalization version 9 is not implemented.

[lookup1980]

Describe the bug

I got an error when converting a model to TensorFlow.

To Reproduce

What I am doing is:

1. A PyTorch model is converted to ONNX model (PyTorch version is 1.0.0 GPU version):
   torch.onnx.export(model, img, "model.onnx", verbose=True)

2. Loaded the ONNX model and verified it:
   onnx.checker.check_model(model)

3. ONNX model is converted to TensorFlow model:
   onnx-tf convert -t tf -i model.onnx -o model.pb

The error message is:
NotImplementedError: BatchNormalization version 9 is not implemented.

ONNX model file

Can't share the model.

Python, ONNX, ONNX-TF, Tensorflow version
Python version:
2.7.15
ONNX version:
1.3.0, pulled today and built from source
ONNX-TF version:
1.2.1, pip install onnx-tf
Tensorflow version:
1.12.0, GPU version

[ikasnep]

I am having the same problem on
python 3.6.6
tf 1.11.0
onnx and onnx-tf the same as above

[hyenal]

@ikasnep are you also using PyTorch 1.0.0 ? I have the same issue having saved a model in this version of Torch with tensorflow 1.5.0 and python 2.7.5.

[ikasnep]

Yes, PyTorch 1.0.0

[shridharkini6]

Yes ,
tensorflow-gpu==1.5.0
onnx==1.4.1
python 2.7.12
onnx-tf installed from source

[tjingrant]

Hi, I've noticed this issue and will provide a patch to resolve in either this weekend or early next week.

[shridharkini6]

@tjingrant any updates on this...

[tjingrant]

@shridharkini6 it's in for review (#361), should be checked in soon.

[davidenitti]

@tjingrant thank you for the fix, is it possible to fix also Cast version 9 is not implemented.?
thanks

[tjingrant]

@davidenitti sure thing. Would you create another issue for tracking purpose?

[davidenitti]

@davidenitti sure thing. Would you create another issue for tracking purpose?

sure @tjingrant , I made it here #362
thanks

[lookup1980]

Tried the tip of master, and it works!
I'll close this issue.
Thank you!

[akshayy29]

Tried the tip of master, and it works!
I'll close this issue.
Thank you!

hey! what did you do? I've got the same problem.

[akshayy29]

Describe the bug

I got an error when converting a model to TensorFlow.

To Reproduce

What I am doing is:

1. A PyTorch model is converted to ONNX model (PyTorch version is 1.0.0 GPU version):
   torch.onnx.export(model, img, "model.onnx", verbose=True)
2. Loaded the ONNX model and verified it:
   onnx.checker.check_model(model)
3. ONNX model is converted to TensorFlow model:
   onnx-tf convert -t tf -i model.onnx -o model.pb

The error message is:
NotImplementedError: BatchNormalization version 9 is not implemented.

ONNX model file

Can't share the model.

Python, ONNX, ONNX-TF, Tensorflow version
Python version:
2.7.15
ONNX version:
1.3.0, pulled today and built from source
ONNX-TF version:
1.2.1, pip install onnx-tf
Tensorflow version:
1.12.0, GPU version

hey! what did you do? I've got the same problem.

[jerpint]

I had the same problem, managed to fix it by installing from source (the fix probably isn't on pip yet).

To do this:

pip uninstall onnx-tf
cd && git clone git@github.com:onnx/onnx-tensorflow.git && cd onnx-tensorflow
pip install -e .