I currently using the license-plate-recognition-barrier-0007 from the open model zoo. I've converted the model in FP16 Precision to run it on a Neural Compute Stick. But when I run it the result seem to be incoherent so i try to run on the CPU and the result was right. I compare the result of each layer and it seem that it's the layer named "transpose" which have the type "Permute" that is not doing his job well. I don't understand what happening because the previous Permute layer in the network seem working well.
Do anyone have any idea ?
I'm using OpenVino 2019.2.275.
My code is compile and run on Windows 10 using Visual Studio 2015.
path_to_model: is the path to the license-plate-recognition-barrier-0007 model that I have converted to IR with FP16 precision.
path_to_image: the path to the image to process.
output_to_compare: the layer where I want to compare the outputs.
Thank you for providing that information.
Can you also tell me what command you used to convert the model? Can you also please try running on our sample application to see if you are able to get accurate results on the CPU and Myriad.