Intel® Distribution of OpenVINO™ Toolkit
Community assistance about the Intel® Distribution of OpenVINO™ toolkit, OpenCV, and all aspects of computer vision-related on Intel® platforms.
6392 Discussions

NCS support for multidimensional output

idata
Employee
537 Views

Hi there,

 

Does anyone have any examples of how to use the Tensor Descriptor class documented here on fifo output:

 

https://movidius.github.io/ncsdk/ncapi/ncapi2/py_api/TensorDescriptor.html

 

I'm currently getting output from a model as a single row vector of 1,082,109, but I need it to be shaped as 1 x 3 x 227 x 227

 

Many thanks..

0 Kudos
8 Replies
idata
Employee
283 Views

@Jay What model are you using? The output from a model is usually not equal to the input dimensions. Like with gendernet, the output is a tensor of length 2, one for each category of the model (male/female).

0 Kudos
idata
Employee
283 Views

Hi @Tome_at_Intel, its a simple segmentation model based on Alexnet. Currently getting the correct number of values back, but they appear to be quite different with a version run on the PC. I didn't manually configure the Tensor Descriptor class in the end, but just used a simple numpy reshape. Let me know if you can think of any issues that might crop up! Thanks!

0 Kudos
idata
Employee
283 Views

@Jay I see. Sorry if I misunderstood you. Is there any way you can provide me a link to the model for testing? Thanks.

0 Kudos
idata
Employee
283 Views

Hi @Tome_at_Intel,

 

The prototxt file is here: https://1drv.ms/u/s!AosBHN3IHTLWhoBSD20uV4cQcxAH5A

 

The weights file is here: http://dl.caffe.berkeleyvision.org/fcn-alexnet-pascal.caffemodel

 

Still having some difficulty matching the PC and NCS output values, so any help appreciated!

 

Many thanks..

0 Kudos
idata
Employee
283 Views

@Jay What kind of results were you seeing? Using mvNCCheck, I received the following results (mvNCCheck runs an inference on the NCS and on the framework and compares the two):

 

mvNCCheck AlexDeploy.prototxt -w fcn-alexnet-pascal.caffemodel - s 12 /usr/lib/python3/dist-packages/scipy/stats/morestats.py:16: DeprecationWarning: Importing from numpy.testing.decorators is deprecated, import from numpy.testing instead. from numpy.testing.decorators import setastest /usr/lib/python3.5/importlib/_bootstrap.py:222: RuntimeWarning: numpy.dtype size changed, may indicate binary incompatibility. Expected 96, got 88 return f(*args, **kwds) /usr/local/bin/ncsdk/Controllers/Parsers/TensorFlowParser/Convolution.py:44: SyntaxWarning: assertion is always true, perhaps remove parentheses? assert(False, "Layer type not supported by Convolution: " + obj.type) mvNCCheck v02.00, Copyright @ Intel Corporation 2017 /usr/local/bin/ncsdk/Controllers/FileIO.py:65: UserWarning: You are using a large type. Consider reducing your data sizes for best performance Blob generated USB: Transferring Data... /usr/local/lib/python3.5/dist-packages/mvnc/mvncapi.py:420: DeprecationWarning: The binary mode of fromstring is deprecated, as it behaves surprisingly on unicode inputs. Use frombuffer instead USB: Myriad Execution Finished USB: Myriad Connection Closing. USB: Myriad Connection Closed. Result: (21, 227, 227) 1) 24852 2.338 2) 24851 2.332 3) 24853 2.332 4) 25079 2.328 5) 24625 2.328 Expected: (21, 227, 227) 1) 24852 2.338 2) 24853 2.332 3) 24851 2.332 4) 24625 2.328 5) 25079 2.328 ------------------------------------------------------------ Obtained values ------------------------------------------------------------ Obtained Min Pixel Accuracy: 0.41771093383431435% (max allowed=2%), Pass Obtained Average Pixel Accuracy: 0.041398050962015986% (max allowed=1%), Pass Obtained Percentage of wrong values: 0.0% (max allowed=0%), Pass Obtained Pixel-wise L2 error: 0.057202097278205216% (max allowed=1%), Pass Obtained Global Sum Difference: 1047.3095703125 ------------------------------------------------------------

 

It seems like the NCS values (Result) are matching the Caffe ones (Expected). Can you provide your python app for further debugging?

0 Kudos
idata
Employee
283 Views

Thanks for the update @Tome_at_Intel

 

Sent you on links there for further investigation..

0 Kudos
idata
Employee
283 Views

Any update for this @Tome_at_Intel

 

Heard nothing back from PM..

0 Kudos
idata
Employee
283 Views

@Jay Still looking into your request. I'll update you as soon as I find something. Thanks for your patience.

0 Kudos
Reply