Fast RCNN implementation on NCS2

[ATell6]

Hello all;

I have a fasterRCNN and a fastRCNN gesture detectors implemented in MATLAB. I have only been successful exporting the fastRCNN network to ONNX.

I would like to use this ONNX model in NCS2.

I only see implementations of faster RCNN detectors as examples.

Could anyone provide an example of Fast RCNN detector for NCS2, where I can use my ONNX model after importing it to the format expected by the Intel device?

Thank you all in advance,

Alberto

[RandallMan_B_Intel]

Hello, ATell6,   

The FastRCNN has been deprecated. 

We have the object detection FasterRCNN demo here: https://github.com/opencv/open_model_zoo/tree/master/demos/object_detection_demo_faster_rcnn  

Please consider that ONNX is not supported by the MYRIAD plugin in OpenVINO™, you would need to use one of the networks listed here:  

https://docs.openvinotoolkit.org/latest/_docs_IE_DG_supported_plugins_MYRIAD.html  

Best Regards,    

Randall B.    

Intel Customer Support Technician    

A Contingent Worker at Intel 

[ATell6]

Thank you for your quick response. This is my first project with NCS2, so I still have some doubts:

1.- I had already seen the example. Is it not possible to implement Fast RCNN? Faster RCNN is ok also for me, but I need to use a network I have created.

2.- In Faster RCNN, is it not possible to use a personally created network? Do I have to stick to the predefined networks supported?

3.- If ONNX format is not supported, how I could I use a network which has been trained in another framework? Must I use TensorFlow or Caffe?

Thanks again in advance, best regards,

Alberto

[RandallMan_B_Intel]

Hello ATell6, 

Thank you for your response, sure.

Yes, you can train your own Faster RCNN, but with the Intel® Neural Compute Stick 2 (NCS2), you would need to use a Tensorflow "faster_rcnn_inception_v2", "faster_rcnn_resnet101". 

If you look at the .json files used to convert, there are faster_rcnn_support_api_v1.*.json to match the TensorFlow version you used to train. You can check them here:  

https://docs.openvinotoolkit.org/latest/_docs_MO_DG_prepare_model_convert_model_tf_specific_Convert_Object_Detection_API_Models.html  

 If you have more questions, please feel free to ask us, and we will help you further. 

Best Regards,    

Randall B.    

Intel Customer Support Technician    

A Contingent Worker at Intel 

[RandallMan_B_Intel]

Hello ATell6,

I have not received a response from you. I would like to know if the issue was solved.

If you have more questions, please feel free to contact us back for further assistance.

Best Regards,    

Randall B.    

Intel Customer Support Technician    

A Contingent Worker at Intel

[ATell6]

Hi,

Due to other obligations at work, I have not had the opportunity to fully test your suggestion.

As I can read from your response, do I have to attach to these architectures?

I use MATLAB for model generation and training. Is there any way to use them in the NCS2?

Anyway, I will test the example you provide.

If I have more questions/doubts, I will contact you again.

Thank you very much for your help

Best,

Alberto

[RandallMan_B_Intel]

Hello ATell6,

I understand.   

Yes, only these topologies.   

Matlab is not supported by the NCS2.   

We don't have a guide on converting MATLAB models with OpenVINO; however, if there is a way MATLAB can export the model in one of the formats OpenVINO supports, it may be possible but we have not tried it.

Please check the MYRIAD Plugin documentation here for supported topologies:  https://docs.openvinotoolkit.org/latest/_docs_IE_DG_supported_plugins_MYRIAD.html  

Sure, if you have more questions, I will be glad to assist you. 

Best Regards,    

Randall B.    

Intel Customer Support Technician    

A Contingent Worker at Intel 

[RandallMan_B_Intel]

Hello ATell6,

I will check on this thread until next Friday. In case you need further assistance I will be happy to help you.

Best Regards,    

Randall B.    

Intel Customer Support Technician    

A Contingent Worker at Intel