NotImplementedError: StridedSlice version 1 is not implemented.

[yjcn]

While I try to export a tensorflow model to onnx based on the suggestions in page: https://github.com/onnx/tutorials/blob/master/tutorials/OnnxTensorflowExport.ipynb I use onnx-tf convert -t onnx -i textdetection.pb -o txtd.onnx However there is an error NotImplementedError: StridedSlice version 1 is not implemented. And I install onnx-tf from source Does anyone know how to solve it?

[tjingrant]

Would you care to share your tf model with us?

[tjingrant]

One other piece of information that would be helpful to us is your onnx version. You can run this script for us to collect version information about your working environment : https://github.com/onnx/onnx-tensorflow/blob/master/util/get_version.py .

StridedSlice is implemented, but it is implemented for version 9 and onward. So somehow your version of onnx-tf was restricted to use a low opset or fails to find the proper handler. Either way, the solution should be rather straightforward.

[yjcn]

Would you care to share your tf model with us?

Of course, This is my tf model. The output node name is "feature_fusion/Conv_7/Sigmoid" and "feature_fusion/concat_3", the input node name is "input_images" This is the version info.

Python version:
3.6.5 |Anaconda, Inc.| (default, Mar 29 2018, 13:32:41) [MSC v.1900 64 bit (AMD64)]
ONNX version:
1.3.0
ONNX-TF version:
1.2.0
Tensorflow version:
1.10.0

[tjingrant]

Hi @yjcn thanks for your response, so it turns out that to support strided slice, we used an experimental onnx op (DynamicSlice), which requires you to install onnx from source as well. You must have installed a release version of onnx (likely you used pip install).

I also ran our model checker on your model and here's the result:

onnx-tf check -i textdetection.pb
INFO:root:Checking for unsupported operators...
INFO:root:There are 18 unique operators in the model file.
INFO:root:There are 1 operators currently not supported by the ONNX-Tensorflow frontend for your model.
INFO:root:set([u'ResizeBilinear'])
INFO:root:Checking for required node arguments...
INFO:root:We checked 549 supported nodes for required arguments.
INFO:root:  # of nodes passed the args check: 549
INFO:root:  # of nodes failed the args check: 0
INFO:root:Some work is needed before we can export your model to ONNX.

Some work is required, but definitely very promising to get it fully working. Do you think you can get rid of this resize bilinear op? Or represent it using existing onnx ops? We can help you out by updating our frontend if you let us know how would you map this resize bilinear op.

[yjcn]

@tjingrant The ResizeBilinear op is from tf.image.resize_bilinear.And I use it to accomplish the Unpooling op.

[tjingrant]

@yjcn sounds like we could map it to onnx Upsample op. I'll get back to you tomorrow, possibly with a patch to do so.

[dolmens]

Still suffering this problem

[tjingrant]

@dolmens please install onnx from source if your onnx-tf doesn't find StridedSlice.

Otherwise, I'm working on the patch (https://github.com/onnx/onnx-tensorflow/pull/333/files).

[pxEkin]

While I try to export a tensorflow model to onnx based on the suggestions in page: https://github.com/onnx/tutorials/blob/master/tutorials/OnnxTensorflowExport.ipynb I use onnx-tf convert -t onnx -i textdetection.pb -o txtd.onnx However there is an error NotImplementedError: StridedSlice version 1 is not implemented. And I install onnx-tf from source Does anyone know how to solve it?

@yjcn
Hi，yjcn， Is the problem solved？ How deal with it？

[iamshri8]

I have converted my Keras model to .pb file and trying to export to onnx, while i am getting the following error.

Could someone help in resolving this?

I installed onnx from pip.

[NicolasHorMann]

Would you care to share your tf model with us?

Of course, This is my tf model. The output node name is "feature_fusion/Conv_7/Sigmoid" and "feature_fusion/concat_3", the input node name is "input_images" This is the version info.

Python version:
3.6.5 |Anaconda, Inc.| (default, Mar 29 2018, 13:32:41) [MSC v.1900 64 bit (AMD64)]
ONNX version:
1.3.0
ONNX-TF version:
1.2.0
Tensorflow version:
1.10.0

Same. I tried with that model.pb and I got this:

I installed all from pip.

[fumihwh]

@NicolasHorMann from source or PyPi?

[NicolasHorMann]

@fumihwh PyPi. I'm using jupyterlab

[fumihwh]

@NicolasHorMann @iamshri8 StridedSlice (DynamicSlice) is added from opset 9, which is 8 in ONNX 1.3.0 (from PyPi). So, currently, StridedSlice does't work with ONNX installed from PyPi. You could press them to release a new version.

[dingzeyuli]

@fumihwh On my mac, I complied ONNX from source and installed tensorflow and onnx-tf from pip3.

Python version:
3.7.1 (default, Nov  6 2018, 18:45:35) 
[Clang 10.0.0 (clang-1000.11.45.5)]
ONNX version:
1.4.1
ONNX-TF version:
1.2.1
Tensorflow version:
1.13.1

I still get the error: NotImplementedError: StridedSlice version 1 is not implemented.

Do you know possible reasons? Thank you.