Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.
6392 Discussions

Re: Error while converting PMTD ONNX model to OpenVino IR

Sharma__Divyanshu
New Contributor I
2,512 Views

Continuation of: https://community.intel.com/t5/Intel-Distribution-of-OpenVINO/Custom-Implementation-for-ScatterND-op/m-p/1201227#M20318 since it is no longer being monitored. 

 

Hi @Max_L_Intel and @Munesh_Intel !

I get the same error despite building the latest version of the model_optimizer from GitHub. I have attached the error log below

 

Thanks a lot for helping with this. It is kind of urgent.

 
 
0 Kudos
1 Solution
Max_L_Intel
Moderator
2,481 Views

Hi @Sharma__Divyanshu 

We tested your model with OpenVINO Github release, but we don't see errors related to ScatterND. The support for ScatterND should be already implemented.

However, there is different error related to unsupported Resize-11 operation:

Model Optimizer arguments:
Common parameters:
	- Path to the Input Model: 	/home/user/Downloads/PMTD_FINAL 1.onnx
	- Path for generated IR: 	/home/user/Downloads/openvino_new/openvino/model-optimizer/.
	- IR output name: 	PMTD_FINAL 1
	- Log level: 	ERROR
	- Batch: 	Not specified, inherited from the model
	- Input layers: 	Not specified, inherited from the model
	- Output layers: 	Not specified, inherited from the model
	- Input shapes: 	Not specified, inherited from the model
	- Mean values: 	Not specified
	- Scale values: 	Not specified
	- Scale factor: 	Not specified
	- Precision of IR: 	FP16
	- Enable fusing: 	True
	- Enable grouped convolutions fusing: 	True
	- Move mean values to preprocess section: 	False
	- Reverse input channels: 	False
ONNX specific parameters:
Model Optimizer version: 	unknown version
[ ERROR ]  Exception occurred during running replacer "REPLACEMENT_ID" (<class 'extensions.load.onnx.loader.ONNXLoader'>): Unexpected exception happened during extracting attributes for node Resize_175.
Original exception message: ONNX Resize operation from opset 11 is not supported.


Resize Opset-11 operation is currently not supported by Model Optimizer to convert. 
OpenVINO team is working on adding support for Resize-11. There are some PRs related to this already: operation specification and support for the operation in the Model Optimizer
As possible workaround you can try to use other Resize-like operation and convert the model with Opset-10.

View solution in original post

0 Kudos
7 Replies
Max_L_Intel
Moderator
2,508 Views

Hi @Sharma__Divyanshu 

Can you please share your model with us, so we try to reproduce this from our end?

Please also note the is being new feature that has been just implemented but not yet verified, hence for the time being we can't guarantee this works as expected. 
Thanks.

0 Kudos
Sharma__Divyanshu
New Contributor I
2,502 Views

Hi @Max_L_Intel !

Do you have an email ID I could send it on. I don't think I'm allowed to post it on a public forum.

Thanks!

0 Kudos
Max_L_Intel
Moderator
2,490 Views

Hi @Sharma__Divyanshu 
We have sent you an email. Please check it.

0 Kudos
Max_L_Intel
Moderator
2,482 Views

Hi @Sharma__Divyanshu 

We tested your model with OpenVINO Github release, but we don't see errors related to ScatterND. The support for ScatterND should be already implemented.

However, there is different error related to unsupported Resize-11 operation:

Model Optimizer arguments:
Common parameters:
	- Path to the Input Model: 	/home/user/Downloads/PMTD_FINAL 1.onnx
	- Path for generated IR: 	/home/user/Downloads/openvino_new/openvino/model-optimizer/.
	- IR output name: 	PMTD_FINAL 1
	- Log level: 	ERROR
	- Batch: 	Not specified, inherited from the model
	- Input layers: 	Not specified, inherited from the model
	- Output layers: 	Not specified, inherited from the model
	- Input shapes: 	Not specified, inherited from the model
	- Mean values: 	Not specified
	- Scale values: 	Not specified
	- Scale factor: 	Not specified
	- Precision of IR: 	FP16
	- Enable fusing: 	True
	- Enable grouped convolutions fusing: 	True
	- Move mean values to preprocess section: 	False
	- Reverse input channels: 	False
ONNX specific parameters:
Model Optimizer version: 	unknown version
[ ERROR ]  Exception occurred during running replacer "REPLACEMENT_ID" (<class 'extensions.load.onnx.loader.ONNXLoader'>): Unexpected exception happened during extracting attributes for node Resize_175.
Original exception message: ONNX Resize operation from opset 11 is not supported.


Resize Opset-11 operation is currently not supported by Model Optimizer to convert. 
OpenVINO team is working on adding support for Resize-11. There are some PRs related to this already: operation specification and support for the operation in the Model Optimizer
As possible workaround you can try to use other Resize-like operation and convert the model with Opset-10.

0 Kudos
Sharma__Divyanshu
New Contributor I
2,473 Views

Hi @Max_L_Intel ! Thanks for checking out the model. I can't use Opset 10. I get the error:

 

RuntimeError: Unsupported: ONNX export of index_put in opset 9. Please try opset version 11.

 

Anyway, since my entire model only requires an upscale factor of 2, I managed to get Resize working for Opset 11 by hardcoding the output shape (By editing the Upsample.py). Once that was done, I commented the code which throws the exception for Resize not being supported in Opset 11. I've successfully tested that this works in a model I converted before (Compared output of PyTorch and OpenVino model).

Attached below are the files I changed. Perhaps you could try converting the model with these files. I get the ScatterND error once this is done.

1. openvino\model-optimizer\extensions\front\onnx\resize_ext.py

2. openvino\model-optimizer\extensions\ops\upsample.py

Thanks a lot!! 

0 Kudos
Max_L_Intel
Moderator
2,460 Views

Hi @Sharma__Divyanshu 

I don't think we could support such method. We tested the files you attached, and we see the same error.

I recommend to wait until Resize-11 support is implemented from OpenVINO side. And please also note that ScatterND support is new feature that has been implemented just recently but not yet fully verified, hence for the time being we can't guarantee this works as expected. 

We apologize for the inconvenience. 

0 Kudos
Max_L_Intel
Moderator
2,432 Views

If you need any additional information from Intel, please submit a new question as this thread is no longer being monitored.

0 Kudos
Reply