Re: mvNCCompile _caffe Error

idata · ‎04-23-2019

I'm using tensorflow and use the command mvNCCompile TF_Model/tf_model.meta -in=conv2d_1_input -on=dense_2/Softmax but got this error.

ImportError: No module named 'numpy.core._multiarray_umath'
Traceback (most recent call last):
  File "/usr/local/bin/mvNCCompile", line 35, in <module>
    from Controllers.Scheduler import load_myriad_config, load_network
  File "/usr/local/bin/ncsdk/Controllers/Scheduler.py", line 22, in <module>
    from Models.Network import *
  File "/usr/local/bin/ncsdk/Models/Network.py", line 18, in <module>
    from Controllers.NCE import NCE_Scheduler
  File "/usr/local/bin/ncsdk/Controllers/NCE.py", line 26, in <module>
    from Models.NetworkStage import NetworkStage
  File "/usr/local/bin/ncsdk/Models/NetworkStage.py", line 24, in <module>
    from Controllers.CnnHardware import *
  File "/usr/local/bin/ncsdk/Controllers/CnnHardware.py", line 24, in <module>
    from Models.StageDefinitions.OpManager import get_op_definition
  File "/usr/local/bin/ncsdk/Models/StageDefinitions/OpManager.py", line 23, in <module>
    from Models.StageDefinitions.Convolution import *
  File "/usr/local/bin/ncsdk/Models/StageDefinitions/Convolution.py", line 22, in <module>
    import Models.StageDefinitions.OpHelpers as OpHelpers
  File "/usr/local/bin/ncsdk/Models/StageDefinitions/OpHelpers.py", line 17, in <module>
    import Controllers.Parsers.Parser.Convolution2D as Layer
  File "/usr/local/bin/ncsdk/Controllers/Parsers/__init__.py", line 3, in <module>
    from .Caffe import CaffeParser
  File "/usr/local/bin/ncsdk/Controllers/Parsers/Caffe.py", line 30, in <module>
    import caffe
  File "/opt/movidius/caffe/python/caffe/__init__.py", line 1, in <module>
    from .pycaffe import Net, SGDSolver, NesterovSolver, AdaGradSolver, RMSPropSolver, AdaDeltaSolver, AdamSolver
  File "/opt/movidius/caffe/python/caffe/pycaffe.py", line 13, in <module>
    from ._caffe import Net, SGDSolver, NesterovSolver, AdaGradSolver, \
SystemError: initialization of _caffe raised unreported exception

Is there something to do with caffe even though I didn't mean to use it?

idata · ‎04-23-2019

Numpy version is 1.15.

There's _validate_lengths not found in 1.16.

idata · ‎04-23-2019

Hi @DableUTeeF

I think this is caused by mixed versions of Numpy installed. Have you tried uninstalling then reinstalling Numpy? Are you getting a "cannot import name '_validate_lengths_' error with v1.16? This Github thread might be helpful to you. Please let me know if this worked.

Best Regards,

Sahira

idata · ‎04-25-2019

It's actually something to do with scipy, I need numpy 1.16 for both Tensorflow and Caffe, so I need to use the newest version of scipy as well, compiling the wheel take 3 hours.

I didn't notice but does the ncsdk compiler downgrade the scipy?

It work without a problem before I install it though.