some questions about NCSDKv2

[idata]

@neal_at_intel @Tome_at_Intel

I notice that NCSDKv2 has been released last Friday. The most significant change between v2 and v1 is introducing the FIFO, and one device can support up to 10 graphs. I went through the API docs and have some questions need your help.

1. Assuming one NCS device can support 10 graphs, those graphs should be the same type or can be different types, such as GoogleNet, AlexNet on one device.
2. What's the relationship of FIFOs and graphs, 1:1, 1:N, N:1, or N:N? In another word, is it possible that one graph reads tensors from more than one inFIFO and write results to more than one outFIFO, vise versa?

Thanks.

[idata]

@xhuan28 NCSDK v2.xx allows for the allocation of different graph files from different models on a single NCS device. You can then queue up inferences from any of these loaded graph files. Typically it is one input FIFO and one output FIFO for every graph, however you can have more than one graph per FIFO if the associated models have the same input/output shape. You can read more about this at: https://movidius.github.io/ncsdk/ncapi/ncapi2/py_api/Fifo.html.

edit: clarification for more than one graph per fifo