Problem compiling my tensorflow model

[FSern]

Hello,

I'm trying to compile a tensorflow model (ssd mobilenet) to mount on the raspberry+NCS, but I'm unable to compile it.

the model I have its the ssd_mobilenet_v1_coco from the tensorflow api (https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/detection_model_zoo.md).

To compile it I tried this command:

mvNCCompile model.ckpt.meta -s 12 -is 300 300 -o ssd_mobilenet.graph

but I get this error:

KeyError: "The name 'input:0' refers to a Tensor which does not exist. The operation, 'input', does not exist in the graph."

so I looked for the input and output nodes (which I'm not sure are correct) and used this command:

mvNCCompile model.ckpt.meta -in=image_tensor -on=detection_boxes -s 12 -is 300 300 -o ssd_mobilenet.graph

but now I get this error

input_data = np.random.uniform(0, 1, shape)File "mtrand.pyx", line 1307, in mtrand.RandomState.uniformFile "mtrand.pyx", line 242, in mtrand.cont2_array_scTypeError: 'NoneType' object cannot be interpreted as an integer

I'm not sure what should I do to be able to compile the model, or even if the nodes are the correct ones. I got the nodes from this graph

[JAVIERJOSE_A_Intel]

Hi FSern,

Thank you for reaching out.

As you are trying to run a TensorFlow SSD MobileNET Network with the Intel® Movidius™ Neural Compute SDK, you will need to create a new config file and use the "--tf-ssd-config" option with the associated command.

You can find the information about how to use the mvNCCompile tool here and how to set the different parameters of the TensorFlow SSD config file here. You need to keep in mind that parameter it's only available on the latest Intel® Movidius™ NCSDK 2.10.

On the other hand, the Intel® Movidius™ NCSDK is no longer being maintained, so, we recommend you migrate your project to the OpenVINO™ toolkit. 

Regards,

Javier A.  

Intel Customer Support Technician  

A Contingent Worker at Intel 

[FSern]

thank you, I was able to compile it on my computer using ncsdk 2 and the flag you mentioned.

I can't run it on the device yet cause I have ncsdk 1.12 on the raspberry so I will have to upgrade it first.

About openvino, can I make it work on the same movidius device? because in that case I maybe should just skip installing the ncsdk 2 and just try to make it work with openvino

[JAVIERJOSE_A_Intel]

Hi FSern,

I am glad to hear you were able to compile the model.

Sure, you can make it work on the Intel® Movidius™ Neural Compute Stick and the Intel® Neural Compute Stick 2 using OpenVINO™ toolkit.

Please read the guide to switch from the Intel® Movidius™ Neural Compute SDK (NCSDK) to the Intel® Distribution of OpenVINO™ Toolkit here.

Regards,

Javier A.  

Intel Customer Support Technician  

A Contingent Worker at Intel