No module named _ext in dcn_v2.py

[heartInsert]

When I successful sh 'make.sh', I got this import error , is there anyone who have met the same problem before?

[heartInsert]

@CharlesShang Thanks for reply

[heartInsert]

By the way , my pytorch version is 1.1.0 , does it affects the result that should I dowdgrade it to 1.0.0?

[maminglang1991]

me too...

[zzzzwj]

Me too. When I run make.sh on Ubuntu 18.04 with cuda 9.0, it worked while run on Ubuntu 16.04 with cuda 7.5, it failed. I'm confused about it. I also met another problem: ImportError:/home/xxx/DCNv2/_ext.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN2at19UndefinedTensorImpl10_singletonE.

[kevin-collins]

When I successful sh 'make.sh', I got this import error , is there anyone who have met the same problem before?

Can you tell me how do you successful sh 'make.sh'?need special configuration?and are you running it under Window10_64?

[heartInsert]

When I successful sh 'make.sh', I got this import error , is there anyone who have met the same problem before?

Can you tell me how do you successful sh 'make.sh'?need special configuration?and are you running it under Window10_64?

Yes,I run it in window10_64, you must change some configuration,
https://github.com/xingyizhou/CenterNet/issues/7 fllow this issues , look carefully

[kevin-collins]

When I successful sh 'make.sh', I got this import error , is there anyone who have met the same problem before?

Can you tell me how do you successful sh 'make.sh'?need special configuration?and are you running it under Window10_64?

Yes,I run it in window10_64, you must change some configuration, https://github.com/xingyizhou/CenterNet/issues/7 fllow this issues , look carefully

Thanks！！！

[SkyeLu]

Has any one solved it? I met the same problem.

[Fighting-JJ]

python 3.6 works well

[ntoxeg]

I just invoked python setup.py install (in a Conda environment) and this solved the problem.

[huohuotm]

ImportError:/home/xxx/DCNv2/_ext.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN2at19UndefinedTensorImpl10_singletonE.

@zzzzwj did you fix this problem?

[zzzzwj]

ImportError:/home/xxx/DCNv2/_ext.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZN2at19UndefinedTensorImpl10_singletonE.

@zzzzwj did you fix this problem?

Sorry, I didn't. I switched to another machine after that. But when I try to run it again, it seems work now. All I did is removing the _ext.cpythonxxxx and recompiling it. But the environment may be not the same as it was when I raised this issue.

[IrisDinge]

i fix this bug by

1. delete whole package
2. shift conda env to pytorch 0.4.0, python 3.6, cuda 9.0, 2080Ti(it doesn't matter I guess)
3. cd DCNv2
4. git checkout pytorch_0.4(then u will check directory has been changed)
5. ./make.sh
6. python test.py
7. run

Good Luck guys;)

[dankydev]

i fix this bug by

1. delete whole package
2. shift conda env to pytorch 0.4.0, python 3.6, cuda 9.0, 2080Ti(it doesn't matter I guess)
3. cd DCNv2
4. git checkout pytorch_0.4(then u will check directory has been changed)
5. ./make.sh
6. python test.py
7. run

Good Luck guys;)

This is not the way it should be solved, since this error is supposed to appear when the user tries to migrate the network to Pytorch >= 1.0. @CharlesShang can you provide a valid and definitive solution please? I followed the instructions, i launched python setup.py install and ./make.sh but it keeps saying ModuleNotFoundError: No module named '_ext'

[shuyu888]

ModuleNotFoundError: No module named '_ext'