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.

Openvino add cpu extension



So I am converting retinanet for object detection, after I convert the model to XML and BIN, I try to use the provided ssd python code to test the converted model.

Here is my code for running the script:

python3 -m openvino_test_model/ssd_v2_for_retina/frozen_inference_graph.xml -i car_1.bmp 

But I got following error:

[ ERROR ] Following layers are not supported by the plugin for specified device CPU:
 Resample_6828, Resample_, PriorBoxClustered_4, PriorBoxClustered_3, PriorBoxClustered_2, PriorBoxClustered_1, PriorBoxClustered_0, DetectionOutput
[ ERROR ] Please try to specify cpu extensions library path in demo's command line parameters using -l or --cpu_extension command line argument

I searched the Document, and find that this should be included in CPU extension lib (i am testing using CPU). However in openvino API I didn't find how and where to utilize this lib.  I searched the name for those layers and find out they are located in deployment_tools/model_optimizer/extensions/ops/. But I dont know how to use them. I assume I should use the -l/--cpu_extension parameter but I dont know what is the argument.



0 Kudos
2 Replies

So, I noticed that the cpu extension in python for -l is in /opt/intel/openvino/deployment_tools/inference_engine/lib/intel64/ 

And also, the is not well written. It only considers the case when async is used. If you are testing with one image, then it wont work. (though the code notation says it can deal with single image) . One needs to modify this code for yourself.


Is there a VPU extension libs?

0 Kudos

Dear Yang, Yuchen,

We have released a new version of OpenVino 2019R1.1 . Please try it, many issues have been fixed. I'm glad you found out about the -l switch. You can find detailed documentation about all OpenVino samples at the doc link below:

Also the  does only use the Async API (as you noted). But we have a non-Async C++ version. The C++ code should be easily translatable to Python.




0 Kudos