Working with the python API (OpenVINO), R5,
We get different results for the same code when replacing the Movidius stick one vs two (rerun the program after replacing the stick, yields different results):
* On a network we converted using mo.py, we compare the output after each layer between Movidius stick one vs two and got considerable differences (for some 100% difference on average). On this example Movidius stick one produces the required results, whereas the Movidius stick two doesn't.
* On an example sample network that came with OpenVINO, we get small differences after each layer between Movidius stick one vs two (no more than 10% difference on average), but the output shape of some layers is different between Movidius stick one vs two (at least when working with the Python interface) (this is not an 'internal layer' we created a new network that ends with the examined layer for each test).
Would appreciate your help:
How can we make the Movidius stick two output be of the same results (output values and shape) as the Movidius stick one?