segmentation fault

[sjf18]

hello, when i am running test.py ,i got an error: segmentation fault, could you help me? thank you!

[sjf18]

when the program runs to 'class _DCNv2(Function):', this error appears

[Mukosame]

Hi this is due to the errors when compiling DCNv2. Can you put more error info after running 'make.sh'?

Also, feel free to check here: https://github.com/CharlesShang/DCNv2/issues/17

[sjf18]

thanks for your reply, accord to this issue, i have updated my gcc version, and compiled dcn sucessfully again, but new error comes,
ImportError: /usr/lib64/libstdc++.so.6: version `GLIBCXX_3.4.20' not found , ImportError: Failed to import DCNv2 module

[Mukosame]

Hi, I find similar errors here: https://github.com/ContinuumIO/anaconda-issues/issues/483. Sorry I'm not an expert in GCC, but people in that repo mentioned that it is because that the lib is only forward compatible.

So my suggestion is, first read their comments and try their solutions. Besides, would you like to provide more detailed information about your environment so that other people might have an idea about how to solve it?

[sjf18]

my system is centos7.3, python version is 3.6.6, pytorch version is 1.1.0, cuda version is 9.2, i have tried to cpmpile dcnv2 sucessfully with gcc version 4.8.5, 4.9.4, 5.5.0, 6.5.0, 7.5.0, but when i use 4.8.5 and 4.9.4, i got segmentation fault error; when i use 5.5.0, 6.5.0, 7.5.0, i got ImportError: /usr/lib64/libstdc++.so.6: version `GLIBCXX_3.4.20' not found , i have seen many issues but i still have no idea how to make dcn work

[Mukosame]

I also failed to install dcn on centos7. Have you tried something like this? Or this?