Opencv vs NCSDK performance

[Pedreira_Gonzalez__I]

Hello i did some test on inference with the same model in openCv and with native SDK, and surprisingly I get a better result with openCv than with the SDK, and I'm talking about a big difference.

Im on a rpi 3b, raspbian stretch, openVino last version and 2019 R3 model

Is that the expected behavior??

Best Regards

[Sahira_Intel]

Hi Ignacio, 

Which model did you use to compare results? Are you able to include your outputs? I would like to verify this on my end if possible. 

Are you using the Neural Compute Stick 2 on the RPI? 

Just to clarify: the NCSDK (mentioned in the subject line) is now unsupported. 

Best Regards,

Sahira

[Pedreira_Gonzalez__I]

Hello

First i have to say sorry as i double posted this and that not my intention, i first send the question and as i do not see published i sent it again.....

Here is the original version, you can find code and model linked on last post
https://software.intel.com/en-us/forums/intel-distribution-of-openvino-toolkit/topic/856367

The model is the one that i downloaded from here
https://download.01.org/opencv/2019/open_model_zoo/R3/20190905_163000_models_bin/

Yes im using Neural Compute Stick 2 on the rpi, and yes sorry i meant openvino inference engine not ncsdk

I did not do anything extrange on my code, only follow code samples and i was surprised to find that openCv is faster than openvino inference_engine.

I even tried asyn version on both approach and still openCv is faster

Maybe im doing something wrong, thats why im asking if this is the expected behaviour

Best Regards