Thought I'd just give you guys a heads up: it seems that the Movidius tools cannot handle small neural networks with less than 8 inputs. See for example the two small nets in this zip file: https://www.dropbox.com/s/yutdhcgof9wuyc5/small_nets.zip?dl=0
This is the output of mvNCCheck for the network with 8 inputs and outputs:
=========================================================
mvNCCheck v02.00, Copyright @ Movidius Ltd 2016
USB: Transferring Data…
USB: Myriad Execution Finished
USB: Myriad Connection Closing.
USB: Myriad Connection Closed.
Result: (1, 8)
1) 0 1.0
2) 7 0.0
3) 6 0.0
4) 5 0.0
5) 4 0.0
Expected: (8,)
1) 5 0.74219
2) 4 0.15564
3) 6 0.094849
4) 3 0.0055466
5) 2 0.00071287
Obtained values
Obtained Min Pixel Accuracy: 134.69732999801636% (max allowed=2%), Fail
Obtained Average Pixel Accuracy: 33.67209732532501% (max allowed=1%), Fail
Obtained Percentage of wrong values: 50.0% (max allowed=0%), Fail
Obtained Pixel-wise L2 error: 59.944784836141054% (max allowed=1%), Fail
Obtained Global Sum Difference: 1.9992806911468506
=========================================================
(Note that the output is not correct, by the way).
This is the result for the network with 7 inputs:
=========================================================
mvNCCheck v02.00, Copyright @ Movidius Ltd 2016
USB: Transferring Data…
[Error 25] Myriad Error: "Status.ERROR".
=========================================================
