can not run "from nvidia.dali.pipeline import Pipeline"

[xxradon]

when I use the 3.1. Binary Installation step [In]( https://docs.nvidia.com/deeplearning/sdk/dali-install-guide/index.html ,everything is ok.But when I 'get started',in python ,run this command" from nvidia.dali.pipeline import Pipeline",I got this error:'ImportError: libdali.so: cannot open shared object file: No such file or directory
'

[JanuszL]

Hi,
Could you tell us which python version do you use?
Can you check status of following command:
pip show nvidia-dali
grab the path from Location:
ls LOCATION/nvidia/dali/.libs/

[cliffwoolley]

I found an error in the RPATH updating in bundle-wheel.sh that likely accounts for this. I have a fix pending.

[cliffwoolley]

(It would still help to know the answers to @JanuszL 's questions so that I can attempt to reproduce the problem. Further: are you using virtualenv or Conda environments or any similar setup?)

[JiangPQ]

Hi, I got a similar problem. Here is the traceback:

Traceback (most recent call last):
File "/home/xxxxx/PycharmProjects/torch90/111.py", line 500, in
from nvidia.dali.pipeline import Pipeline
File "/home/xxxxx/miniconda3/envs/torch90/lib/python3.6/site-packages/nvidia/dali/init.py", line 17, in
from . import ops
File "/home/xxxxx/miniconda3/envs/torch90/lib/python3.6/site-packages/nvidia/dali/ops.py", line 19, in
from nvidia.dali import backend as b
File "/home/xxxxx/miniconda3/envs/torch90/lib/python3.6/site-packages/nvidia/dali/backend.py", line 15, in
from nvidia.dali.backend_impl import *
ImportError: libnppicom.so.9.0: cannot open shared object file: No such file or directory

The lib64 has been added to LD_LIBRARY_PATH and everything goes fine when I run pytorch so i think CUDA9.0 is properly installed.

echo $LD_LIBRARY_PATH
/usr/local/cuda/lib64:/opt/intel/compilers_and_libraries_2018.2.199/linux/ipp/lib/intel64

I use conda environment BTW and this is what pip show nvidia-dali gives:

$ pip show nvidia-dali
Name: nvidia-dali
Version: 0.1.0
Summary: NVIDIA DALI
Home-page: https://github.com/NVIDIA/dali
Author: NVIDIA Corporation
Author-email: UNKNOWN
License: Apache License 2.0
Location: /home/xxxxx/miniconda3/envs/torch90/lib/python3.6/site-packages
Requires: future
Required-by:

@cliffwoolley

[xxradon]

@JanuszL @cliffwoolley
I also use conda environment and this is what pip show nvidia-dali gives:

Name: nvidia-dali
Version: 0.1.1
Summary: NVIDIA DALI
Home-page: https://github.com/NVIDIA/dali
Author: NVIDIA Corporation
Author-email: UNKNOWN
License: Apache License 2.0
Location: /home/shining/ProgramFiles/miniconda/envs/pytorch/lib/python3.6/site-packages
Requires: future

I have search libdali.so in my conda env ,I found nothing.But when I compiled from DALI source,I got the libdali.so.
And @JiangPQ,I think your cuda environment got something wrong,you can test your pytorch cuda to verify.

[JanuszL]

@xxradon
Could you also:
ls /home/shining/ProgramFiles/miniconda/envs/pytorch/lib/python3.6/site-packages/nvidia/dali/.libs/ ?

[cliffwoolley]

It's not called libdali.so in the installed binary package; it's libdali-*.so, where the * is a hash. Bundle-wheel.sh renames it when packaging the wheel. But because of some path errors, not all references to it are corrected properly. I have a fix for that, but the fix exposes another error that I'm still trying to sort out.

[xxradon]

I have test recent release editon,this problem has been fixed.Thank you.

[cliffwoolley]

Actually this isn't fixed yet. :) Glad it worked for you regardless, but I'm reopening this until I push the proper fix.