fail to connect to tf_sample_so.so: tf_sampling_so.so: undefined symbol: _ZN10tensorflow12OpDefBuilder4AttrENS
Opened this issue · 1 comments
we are using tx2 board and these settings: cuda 10.0, cudnn 7.5, tf1.14(also tried tf1.13)
but we failed at the very beginning of train.py
after we compile tf_sampling_so.so and everything seemed just ok, but when we run train.py it just said tf_sampling_so.so: undefined symbol: _ZTIN10tensorflow8OpKernelE at first, so we added -L /usr/local/lib/python3.5/dist-packages/tensorflow -ltensorflow_framework and the the error changed into: undefined symbol: _ZN10tensorflow12OpDefBuilder4AttrENS
we then tried tf1.13 but things still went wrong:(
Is this because the unfitting version of cuda, cudnn, and tensorflow?
Will a downgrade of environment help?
@hhilda Hi, I compile the file
cd ~/EVclouds_gesture_recognition/tf_ops/sampling
$ sh tf_sampling_compile.sh
tf_sampling.cpp:6:10: fatal error: tensorflow/core/framework/op.h: No such file or directory
#include "tensorflow/core/framework/op.h"
^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
compilation terminated.
I use the tensorflow1.0.1, python2.7, gcc7.3.0, cuda11.2. Any suggestions? By the way, do you run the code successfully? Thanks.