@stephen@logic Yes, I can give you a quick overview/rundown. There are some similarities between NCSDK and OpenVINO. Both require the model to be compiled into a binary format. Both have a C++ and Python API.
OpenVINO's Model Optimizer is pretty much the equivalent of NCSDK's mvNCCompile tool. As most of the NCS forum developers know, the mvNCCompile tool creates a Movidius graph file using Caffe/Tensorflow model files. In the same manner, OpenVINO's Model Optimizer creates an Intermediate Representation or IR of the model using the same model files. However, OpenVINO IR and NCSDK graph files are not cross compatible.
When compiling your model with the Model Optimizer, make sure to use the following flag: --data_type=FP16. This will create an IR (Intermediate Representation) that is compatible with the NCS device using half precision floats or FP16.
OpenVINO's Inference Engine is equivalent to the NCSDK's API. This will be how you will perform inferences. When running the examples included with the OpenVINO toolkit, make sure to specify the NCS device using the following flag: -d MYRIAD. For example (if you have built the classification example and compiled Alexnet with the Model Optimizer using FP16 data type), you can run the example with the following command to specify the NCS: ./classification_sample -i cat.bmp -m alexnet_fp16.xml -d MYRIAD. Please feel free to hop over to the OpenVINO forums for more information. Hope this helps!
