Re: ''UnpopulatedTensor" error for running an inference model trained with SSD on Movidius stick

idata · ‎11-02-2018

I have an inference model trained with SSD Caffe.

I put my caffemodel and prototxt files into examples/caffe/GoogleNet folder and change file names to bvlc_googlenet.caffemodel and deploy.prototxt.

So that I don't need to update at different places.

Then inside GoogleNet folder, I run

make run

But I have compile failed errors as follow.

making prereqs
(cd ../../data/ilsvrc12; make)
make[1]: Entering directory '/home/itc/Softwares/Projects/Movidius/ncsdk/examples/data/ilsvrc12'
make[1]: Leaving directory '/home/itc/Softwares/Projects/Movidius/ncsdk/examples/data/ilsvrc12'

making prototxt
Prototxt file already exists

making caffemodel
caffemodel file already exists

making compile
mvNCCompile -w bvlc_googlenet.caffemodel -s 12 deploy.prototxt
/usr/local/bin/ncsdk/Controllers/Parsers/TensorFlowParser/Convolution.py:46: SyntaxWarning: assertion is always true, perhaps remove parentheses?
  assert(False, "Layer type not supported by Convolution: " + obj.type)
/usr/local/bin/ncsdk/Controllers/Parsers/Phases.py:322: SyntaxWarning: assertion is always true, perhaps remove parentheses?
  assert(len(pred) == 1, "Slice not supported to have >1 predecessors")
mvNCCompile v02.00, Copyright @ Intel Corporation 2017

<caffe._caffe.BlobVec object at 0x7fe53b11e800>
<caffe._caffe.BlobVec object at 0x7fe53b11e9e0>
<caffe._caffe.BlobVec object at 0x7fe53b11ebc0>
<caffe._caffe.BlobVec object at 0x7fe53b11eda0>
<caffe._caffe.BlobVec object at 0x7fe53b1225d0>
<caffe._caffe.BlobVec object at 0x7fe53b1227b0>
<caffe._caffe.BlobVec object at 0x7fe53b122990>
<caffe._caffe.BlobVec object at 0x7fe53b122b70>
<caffe._caffe.BlobVec object at 0x7fe53b122d50>
<caffe._caffe.BlobVec object at 0x7fe53b122f30>
<caffe._caffe.BlobVec object at 0x7fe53b114210>
<caffe._caffe.BlobVec object at 0x7fe53b114490>
<caffe._caffe.BlobVec object at 0x7fe53b1146c0>
<caffe._caffe.BlobVec object at 0x7fe53b114940>
<caffe._caffe.BlobVec object at 0x7fe53b114bc0>
<caffe._caffe.BlobVec object at 0x7fe53b114e40>
<caffe._caffe.BlobVec object at 0x7fe53b112120>
<caffe._caffe.BlobVec object at 0x7fe53b112350>
<caffe._caffe.BlobVec object at 0x7fe53b112580>
<caffe._caffe.BlobVec object at 0x7fe53b112760>
<caffe._caffe.BlobVec object at 0x7fe53b1129e0>
<caffe._caffe.BlobVec object at 0x7fe53b112c10>
<caffe._caffe.BlobVec object at 0x7fe53b112e90>
<caffe._caffe.BlobVec object at 0x7fe53b115170>
<caffe._caffe.BlobVec object at 0x7fe53b1153a0>
Traceback (most recent call last):
  File "/usr/local/bin/mvNCCompile", line 206, in <module>
    create_graph(args.network, args.image, args.inputnode, args.outputnode, args.outfile, args.nshaves, args.inputsize, args.weights, args.explicit_concat, args.ma2480, args.scheduler, args.new_parser, args.cpp, args)
  File "/usr/local/bin/mvNCCompile", line 185, in create_graph
    load_ret = load_network(args, parser, myriad_config)
  File "/usr/local/bin/ncsdk/Controllers/Scheduler.py", line 102, in load_network
    parsedLayers = convertBiasLayersToEltwise(parsedLayers)
  File "/usr/local/bin/ncsdk/Controllers/Parsers/Phases.py", line 857, in convertBiasLayersToEltwise
    converted.setInputTensors(first_input)
  File "/usr/local/bin/ncsdk/Controllers/Parsers/Parser/Layer.py", line 112, in setInputTensors
    self.inputTensors = tuple(t)
TypeError: 'UnpopulatedTensor' object is not iterable
Makefile:80: recipe for target 'compile' failed
make: *** [compile] Error 1

What could be wrong?

idata · ‎11-02-2018

@coi can you send me your files? I will try to reproduce the problem.

idata · ‎11-03-2018

@Oliver_at_Intel Hello, thank you.

Please download prototxt file here and

caffemodel is here

idata · ‎11-03-2018

@Oliver_at_Intel Here is labelmap file. Thank you.

idata · ‎11-04-2018

Sorry labelmap file is here.

idata · ‎11-08-2018

@Oliver_at_Intel , hi Oliver! any chance to test my files for the issue. Thanks

idata · ‎11-08-2018

@coi Hi coi, I was able to download and test your model. I was able to reproduce the same issue and I'm still investigating the root cause of the problem. Thanks for your patience. I'll let you know as soon as I find something.

idata · ‎11-08-2018

@coi There seems to be a bug with the NCSDK while parsing your model. Normally the NCSDK takes the Bias layer and changes it to a Eltwise Add, but during this process, an error occurs when setting the input tensor. There isn't a workaround that I can provide at the moment, but I will let you know as soon as we find a solution for this issue. Thanks.

idata · ‎11-09-2018

@Tome_at_Intel Thank you Tome, for your great support. Since I am using the stick for one of my applications, your help will be very grateful for me.

idata · ‎11-22-2018

@Tome_at_Intel Dear Tome, can find a solution for my case?