After I trained my mtcnn pytorch models ( my codes are based on python), I got pnet.pt/rnet.pt/onet.pt and converted them into .onnx files and then into .pb files. After that, I used the Model Optimizer by converting those .pb files into IR files( finally, got pnet.xml and pnet.bin/ rnet.xml and rnet.bin/ onet.xml and onet.bin in FP16).
And.... I was wondering if anyone else could tell me what I should do in the next steps. By the way, my codes are based on python. It is too difficult for me so If you have any constructive advice, please contact me!!
Thank you so much!!!
Solanki, Saurav wrote:
Please find my answer here. Just change the plugin name and if needed add CPU extension and you are good to go.
First, I would like to say thank you to provide me such constructive code. As I am a new one to NCS2 and Python API, I still have some questions and was wondering if it could be possible for you to help me deal with them.
As you can see, now I am using the mtcnn models so I have three nets such as pnet, rnet and onet. So my question is how to load all of them into your Template Code? I mean there is only one place to load IR model:
model1 = "<model_path.%s>" or maybe including(model_xml / model_bin)
Maybe I should write like this?
model1 = "<model_path.%s>"
model2 = "<model_path.%s>"
model3 = "<model_path.%s>"
num_requests = 3
And another question is how I could transfer the size of the picture from the first net (pnet) into the final net (onet)...
From my perspective, it is a big project QAQ....So I was wondering if you could give me some more details about the next steps I need to do.
Thank you so much!!!:)