How does OpenVINO work

[Korolev__Roman]

Hello everyone! I converted my SSD multibox detector from MxNet to IR and found that it works much faster on Intel CPU than just MxNet framework. And that's great! So I would like to have a primitive understanding of how it works. I mean maybe it has some highly optimized conv or pooling operations or something like that. Could anybody explain me in a couple of words? Thanks!

[Luis_at_Intel · Accepted solution]

Hi Korolev, Roman,

Thanks for reaching out. There are quite a few resources that could help understand how optimization works in OpenVINO. As described in the Model Optimizer Guide, there are a number of device-agnostic optimizations the tool performs. For example, certain primitives like linear operations (BatchNorm and ScaleShift), are automatically fused into convolutions. Generally, these layers should not be manifested in the resulting IR (as seen here).

You can find more information about device optimizations for the supported plugins in the Device-Specific Optimizations section of the Optimization Guide. Hope you have found this helpful.

Regards,

Luis