Intel® Distribution of OpenVINO™ Toolkit
Community support and discussions about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all things computer vision-related on Intel® platforms.

CPU plugin not working correctly.

lex96
Beginner
516 Views

Hello, I'm trying to run the SCRFD face detector (https://github.com/deepinsight/insightface/tree/master/detection/scrfd). Wrong outputs are obtained on the CPU (in some cases, random outputs, i.e. with the same input, outputs change from start to start). When launched on the GPU, the outputs are correct (checked for torch_1.6_scrfd_2.5g_bnkps_640_640_opset_version_11.onnx).
OS: Windows
OpenVINO: 2021.2, 2021.4.1.
OpenVINO was installed through the installer from the site.
When exporting to onnx, tried various combinations of pytorch and onnx opset.
What is the problem with the CPU plugin?

Labels (1)
0 Kudos
4 Replies
Iffa_Intel
Moderator
453 Views

Greetings,


Which model and topology are you using?.(eg: model: Tensorflow, Topology: FasterRCNN). Not all models and topologies are supported by OpenVINO.

Plus, which code/sample application did you use for inferencing?



Sincerely,

Iffa



lex96
Beginner
445 Views

Hello, model: Pytorch (ONNX), topology: SCRFD, the used models are provided in the attached archive. I can not provide the code, tk. it is heavily intertwined with the product code. But I successfully use other topologies: RetinaFace, ResNet, ResNext and others.
The problem is that the network works correctly on the GPU plugin.

Iffa_Intel
Moderator
426 Views

Unfortunately, your ONNX model topology is not listed in the Supported Topologies.

Therefore, it might luckily work, but issues are expected.


Both model and topology need to be supported by OpenVINO in order to perform any operations on them (in the perspective of OpenVINO)


Sincerely,

Iffa


Iffa_Intel
Moderator
397 Views

Greetings,


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



Sincerely,

Iffa


Reply