Intel® Compute Stick
Support for Intel® Compute Sticks and Cards
Announcements
Do you have improvements you would like us to make on this community site? If so, we would love to hear your feedback! Click here for an 8 question survey. Thanks!

520 Discussions

what(): Cannot create ShapeOf layer PriorBoxClustered_0/0_port id:403 error in inference NCS2, RPi.

dilip96
New Contributor I
886 Views

Hi ,

I am trying to run inference on my RPi 4 with Rasbian Buster OS and NCS2 setup. The model is custom trained SSD MobileNet V2 model from TF Object detection API. There is no problem in inference using the TF models.

Also the conversion of TF to IR models is successful, without any errors. the command as below.

sudo python3 mo.py --framework tf --input_model /home/rootuser/Workspace/Catering/New_Models/TF/ssd_20k_800.pb --reverse_input_channels --tensorflow_object_detection_api_pipeline_config /home/rootuser/Workspace/Catering/ssd_20k_800.config --tensorflow_use_custom_operations_config extensions/front/tf/ssd_support_api_v1.14.json --output=detection_classes,detection_scores,detection_boxes,num_detections --data_type FP16 --output_dir /home/rootuser/Workspace/Catering/New_Models/ --input_shape [1,300,300,3]

But when I try to run the inference in NCS2 I get the following error.(OpenVINO 2020.3.220 version)

terminate called after throwing an instance of ‘InferenceEngine::details::InferenceEngineException’
  what():  Cannot create ShapeOf layer PriorBoxClustered_0/0_port id:403
Aborted

Couldn't find any help on this error. Please help me out as soon as possible.

 

Note: Update on this issue. I was able to run inference with Model converted using the OpenVINO2019R3.

Is this a bug in 2020 version?

0 Kudos
5 Replies
dilip96
New Contributor I
859 Views

Also on the same issue,

My OpenVino Model does not give me the same performance as my TF models.

Check this link for the previous topic I opened for the same issue. So the suggested solutions are tried.

https://community.intel.com/t5/Intel-Distribution-of-OpenVINO/Model-performance-reduced-not-detectin...

I have attached the Pb files with this. The command to convert my IR is as above.(note: inference result is from IR models converted using OpenVINO 2019.R3.)

 

 

IntelSupport
Community Manager
844 Views

Hi Dilip96,

 

There is a bug. You need to modify the ssd_support_api_v1.14.json file by changing line 57 from "Postprocessor/Cast" to "Postprocessor/Cast_1". Also, in your command, try running the command without giving the --input_shape[1,300,300,3]?

Please try and let me know the result.

 

Regards,

Aznie


dilip96
New Contributor I
823 Views

Hi,

line 57 in my  ssd_support_api_v1.14.json file is already changed to "Postprocessor/Cast_1". 

so that is not the solution for the existing problem.

Also, removing the parameter --input_shape did not give any result. It also give me the same error.

 

terminate called after throwing an instance of ‘InferenceEngine::details::InferenceEngineException’
  what():  Cannot create ShapeOf layer PriorBoxClustered_0/0_port id:403
Aborted

 

Hope I get a solution soon, thank you.

IntelSupport
Community Manager
808 Views

 

Hi Dilip96,

 

Please convert the model using OpenVINO 2021 and use the ssd_support_api_v1.15.json since we have fixed the bug in the latest version. I have tried and able to run the inference using the Object_detection_sample_ssd inference engine.

 

Regards,

Aznie


IntelSupport
Community Manager
778 Views

Hi Dilip96,


This thread will no longer be monitored since we have provided a solution. If you need any additional information from Intel, please submit a new question.


Regards,

Aznie


Reply