As I tried use myriad running example security_barrier_camera_sample on openvino toolkit, it seems performance remain same while I use one or two sticks.
Can I get better performance while having two VPU with any kind of setting?
It does and you should use the R2 release for OpenVINO to test this. What is supported is running N inferences in parallel on N sticks. Not a single inference parallelized across multiple sticks. If you need that then you would have to split your model on the application level. Depending on your application you could get better performance. Are you using more than one model?
Is there any document or example to demonstrate how to do N inferences in parallel on N sticks?
I can run one example on one stick. However, I don't know how to do N inferences in parallel on N sticks.
I want to use two myriad neural computer sticks to run example security_barrier_camera_sample on openVINO toolkit.
Is there any document or example to demonstrate how to do N inferences in parallel on N sticks? Great thanks!
May be you could try this sample
security_barrier_camera_sample.exe -d GPU -d_va MYRIAD -d_lpr MYRIAD -i car07.mp4 -m C:\Intel\computer_vision_sdk_2018.2.317\deployment_tools\intel_models\vehicle-license-plate-detection-barrier-0007\FP16\vehicle-license-plate-detection-barrier-0007.xml -m_va C:\Intel\computer_vision_sdk_2018.2.317\deployment_tools\intel_models\vehicle-attributes-recognition-barrier-0039\FP16\vehicle-attributes-recognition-barrier-0039.xml -m_lpr C:\Intel\computer_vision_sdk_2018.2.317\deployment_tools\intel_models\license-plate-recognition-barrier-0001\FP16\license-plate-recognition-barrier-0001.xml
There is in-package documentation underneath supported devices->myriad that gives you api calls to use in your application to allocate movidius devices for specific networks.
As always, for internal questions there is an inside blue forum for OpenVINO and we would like you to post your questions there.