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.
6528 Discussions

Model Optimizer Error: The ExpandDims node Unsqueeze_333 has more than 1 input

TonyWong
Beginner
2,836 Views

Dear OpenVino Guru,

 

Hi, this is Tony. And I'm back.

Here is my model, retrained and based on Yolo V5.

https://drive.google.com/file/d/14XwO6OHHFDUSKFxhHphNZqlbDIIlEpxS/view?usp=sharing

 

Question 1:

My current model has no input limit for the image height and width. 

Is it possible that I also does not preset the image size in the transformed IR model?

 

Question 2:

Previously, one of your Guru, Aznie, helped me to run the optimizer, but he stopped before the step where error occurs.

And I do need the full model result.(no mid-way.)

 

His command:

$python mo.py --input_model model.onnx --output Conv_410,Conv_322,Conv_498 --input_shape [1,3,512,512]

(I've even twist 512,512 to 480,640, to let the inference step pass)

 

However, using Netron to visualized the model, I've found that the last 1/5 of the model is truncated .

 

Running without   "--output", I've got errors like:


[ ERROR ] The ExpandDims node Unsqueeze_333 has more than 1 input

...........
[ ERROR ] The ExpandDims node Unsqueeze_583 has more than 1 input

 

Please see the attachment for the full output.txt  and the results  with --log_level=DEBUG 

 

Please help me out.

 

With Regards.

Tony

0 Kudos
1 Solution
Zulkifli_Intel
Moderator
2,781 Views

Hello Tony Wong,


My apologies for the previously shared input.


On our replication, we faced the same error as yours when we exclude the --output argument from the MO command. Our findings indicated that our current version is not able to perform unsqueeze function with 2 inputs (data, axes) for the ONNX model.


Our development team is working to rectify this as PR-8399 has been initiated and is pending merging to the master branch.

https://github.com/openvinotoolkit/openvino/pull/8399)


Once merged, you may update your local repository by pulling the master branch using this command:

git pull https://github.com/openvinotoolkit/openvino/tree/master


If you were previously installed OpenVINO using our distribution package, regret to inform that you might need to build the open-source OpenVINO to enjoy this fix.

Temporarily, the current workaround that we would propose is to specify the --input_shape and --output arguments to allow the model to be converted successfully.


Sincerely,

Zulkifli


View solution in original post

9 Replies
Zulkifli_Intel
Moderator
2,804 Views

Hello Tony Wong,

 

Thank you for reaching out to us.

 

I’m able to convert your model using the recommended MO command from the previous thread.

onnx convert.PNG

 

Specifying the input_shape argument is optional. But some models require using this additional argument to specify the conversion parameter. 

 

Sincerely,

Zulkifli

 

0 Kudos
TonyWong
Beginner
2,795 Views

Thanks, Zulkili.

 

Apparently, you did not read my post at all. You are wasting our time.

In the above post, I have explained why former command does not work at all.

 

In short, I want the final result ,not the middle way result.   Not the result before error occurs. 

Please take a look.

 

Thanks.

Tony

 

 

 

 

 

0 Kudos
Zulkifli_Intel
Moderator
2,782 Views

Hello Tony Wong,


My apologies for the previously shared input.


On our replication, we faced the same error as yours when we exclude the --output argument from the MO command. Our findings indicated that our current version is not able to perform unsqueeze function with 2 inputs (data, axes) for the ONNX model.


Our development team is working to rectify this as PR-8399 has been initiated and is pending merging to the master branch.

https://github.com/openvinotoolkit/openvino/pull/8399)


Once merged, you may update your local repository by pulling the master branch using this command:

git pull https://github.com/openvinotoolkit/openvino/tree/master


If you were previously installed OpenVINO using our distribution package, regret to inform that you might need to build the open-source OpenVINO to enjoy this fix.

Temporarily, the current workaround that we would propose is to specify the --input_shape and --output arguments to allow the model to be converted successfully.


Sincerely,

Zulkifli


TonyWong
Beginner
2,751 Views

Many thanks,  Zulkifli.

This is exactly what I'm looking for.

 

Thanks, I will watch this PR on the first moment.

 

Tony

 

0 Kudos
TonyWong
Beginner
2,736 Views

Oh, one tiny small question. 

That PR is there 8 days ago. And not so much code change.

Usually how many days for a PR in Intel , will be reviewed and merged.  

So that I can let my director know the estimation date.

 

Thanks.

 

Tony

 

0 Kudos
Zulkifli_Intel
Moderator
2,708 Views

Hello Tony Wong,


We do not have any specific or estimated time the PR will be merged and released. It all depends on the internal processes of validation, testing, and approval.


Sincerely,

Zulkifli


0 Kudos
Zulkifli_Intel
Moderator
2,631 Views

Hello TonyWong,


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


Sincerely,

Zulkifli


0 Kudos
Zulkifli_Intel
Moderator
2,460 Views

Hello TonyWong,

 

The PR 8399 has been merged, feel free to try with the latest branch. If you faced any issues, please open a new case.

 

Sincerely,

Zulkifli

TonyWong
Beginner
2,456 Views

I have tried, and it works . Thank you, Mr. Zulkifli

 

0 Kudos
Reply