Error Compiling Examples

[idata]

When trying to compile any example using commands like:

making compile

mvNCCompile -w squeezenet_v1.0.caffemodel -s 12 deploy.prototxt

I get the following error:

Makefile:88: recipe for target 'compile' failed

make: *** [compile] Illegal instruction (core dumped)

I have

jose@computer:~/ncsdk/examples/caffe/SqueezeNet$ echo $PYTHONPATH:/opt/movidius/caffe/python

[idata]

@JosePerez What type of machine are you using the NCSDK on?

[idata]

Hi @Tome_at_Intel I am also getting this when the make file for Facenet tries to compile the graph on Ubuntu 16.04 & 18.04 (On an UP Squared). I had seen similar errors with conflicting TF versions before so decided to try upgrade to 18.04 and reinstall everything but no change. ( I modified the NCSDK2 installer to allow 1804)

Running Facenet makefile:

_2018-10-29 13:24:40 (7,19 MB/s) - ‘20170512-110547.zip’ saved [191972345]

making model

Zip file exists.Unzipping.Archive: 20170512-110547.zip inflating: 20170512-110547/20170512-110547.pb inflating: 20170512-110547/model-20170512-110547.ckpt-250000.data-00000-of-00001 inflating: 20170512-110547/model-20170512-110547.ckpt-250000.index inflating: 20170512-110547/model-20170512-110547.metaConverted directory does not exist, doing conversionIllegal instruction (core dumped)

making compile

/bin/sh: 7: cd: can't cd to 20170512-110547/facenet_celeb_ncsCommand line: mvNCCompile facenet_celeb_ncs.meta -w facenet_celeb_ncs -s 12 -in input -on output -o facenet_celeb_ncs.graph/usr/local/lib/python3.6/dist-packages/skimage/init.py:60: DeprecationWarning: the imp module is deprecated in favour of importlib; see the module's documentation for alternative uses import impIllegal instruction (core dumped)cp: cannot stat 'facenet_celeb_ncs.graph': No such file or directory_

[idata]

@AdamMiltonBarker Which CPU does your Up Squared board use? If it is an Atom CPU, it seems like the Intel Atom processors don't have AVX support and for TensorFlow 1.6 and higher, release binaries are prebuilt with AVX instructions. One option is to use TF 1.5 or below.

Another option that may work is to build TensorFlow yourself using docker: https://github.com/hadim/docker-tensorflow-builder

[idata]

Intel NUC NUC5CPYH with 8 Gb of RAM. Running a new installation of Ubuntu 16.04

[idata]

@Tome_at_Intel Will I be able to run in this platform?

[idata]

@JosePerez You should be able to run it with that platform.

[idata]

Any clue to solve this issue? As I mentioned, this is a brand new installation of Ubuntu and then the movidius installation. It is not makeany sence to have any error if we only have this installation from the scratch

[idata]

@JosePerez I was mistaken, the Intel Celeron N3050 processor that comes with the Intel NUC NUC5CPYH doesn't have support for AVX instructions. As such the TensorFlow binaries that were prebuilt with AVX instructions (with pip) won't work on your machine. More information at https://github.com/tensorflow/tensorflow/issues/23520.