InternImage ImportError: .local/lib/python3.8/site-packages/DCNv3-1.0-py3.8-linux-x86_64.egg/DCNv3.cpython-38-x86_64-linux-gnu.so: undefined symbol: _ZN2at4_ops10select_int4callERKNS

Whenever I try to train the model I get the following error:

First, my environment:

NVIDIA-SMI 525.89.02 Driver Version: 525.89.02 CUDA Version: 12.0
NAME="Ubuntu" VERSION="20.04.5 LTS (Focal Fossa)" PYTORCH= 1.14.0a0+44dac51

The full error message:

Traceback (most recent call last):
  File "train.py", line 17, in <module>
    from mmseg.apis import init_random_seed, set_random_seed, train_segmentor
  File "wandb/mmsegmentation/mmseg/apis/__init__.py", line 2, in <module>
    from .inference import inference_segmentor, init_segmentor, show_result_pyplot
  File "wandb/mmsegmentation/mmseg/apis/inference.py", line 9, in <module>
    from mmseg.models import build_segmentor
  File "wandb/mmsegmentation/mmseg/models/__init__.py", line 2, in <module>
    from .backbones import *  # noqa: F401,F403
  File "wandb/mmsegmentation/mmseg/models/backbones/__init__.py", line 23, in <module>
    from .intern_image import InternImage
  File "wandb/mmsegmentation/mmseg/models/backbones/intern_image.py", line 17, in <module>
    from ops_dcnv3 import modules as opsm
  File "wandb/mmsegmentation/ops_dcnv3/modules/__init__.py", line 7, in <module>
    from .dcnv3 import DCNv3, DCNv3_pytorch
  File "wandb/mmsegmentation/ops_dcnv3/modules/dcnv3.py", line 15, in <module>
    from ..functions import DCNv3Function, dcnv3_core_pytorch
  File "mmsegmentation/ops_dcnv3/functions/__init__.py", line 7, in <module>
    from .dcnv3_func import DCNv3Function, dcnv3_core_pytorch
  File "mmsegmentation/ops_dcnv3/functions/dcnv3_func.py", line 17, in <module>
    import DCNv3
ImportError: .local/lib/python3.8/site-packages/DCNv3-1.0-py3.8-linux-x86_64.egg/DCNv3.cpython-38-x86_64-linux-gnu.so: undefined symbol: _ZN2at4_ops10select_int4callERKNS_6TensorEll

Apr 04 '23 09:04 fonzen1

I also encountered this problem before, but it was resolved when I changed the CUDA version to 11.7, torch==1.13.1, torchvision==0.14.1, torchaudio==0.13.1,and mmsegmentation==0.30.0. Maybe you can try it as well. Hope this helps.

Apr 11 '23 04:04 yukiyuki333

I also encountered this problem before, but it was resolved when I changed the CUDA version to 11.7, torch==1.13.1, torchvision==0.14.1, torchaudio==0.13.1,and mmsegmentation==0.30.0. Maybe you can try it as well. Hope this helps.

This works. But I am unable to change CUDA version as I am using a GPU server on the cloud with CUDA 12. Is there any other fix that is compatible with CUDA 12?

Apr 11 '23 07:04 fonzen1

same problem while using torch==1.9.1+cu111, and ./make.sh doesnt work as well

Apr 11 '23 09:04 Italian-PAO

same problem while using torch==1.9.1+cu111, and ./make.sh doesnt work as well

Hi, please install PyTorch>=1.10.0.

Apr 11 '23 11:04 czczup

same problem while using torch==1.9.1+cu111, and ./make.sh doesnt work as well

Hi, please install PyTorch>=1.10.0.

Thank you so much for your quick reply, I have realised this problem and I'm installing PyTorch==1.10.0 Thanks again : )

Apr 11 '23 11:04 Italian-PAO

same problem while using torch==1.10.1+cu111, and ImportError: /~/python3.8/site-packages/DCNv3-1.0-py3.8-linux-x86_64.egg/DCNv3.cpython-38-x86_64-linux-gnu.so: undefined symbol: _ZN2at4_ops5zeros4callEN3c108ArrayRefINS2_6SymIntEEENS2_8optionalINS2_10ScalarTypeEEENS6_INS2_6LayoutEEENS6_INS2_6DeviceEEENS6_IbEE

Apr 17 '23 13:04 Skyninth

same problem while using torch==1.9.1+cu111, and ./make.sh doesnt work as well

Hi, please install PyTorch>=1.10.0.

Thank you so much for your quick reply, I have realised this problem and I'm installing PyTorch==1.10.0 Thanks again : )

Hello, have you installed a higher version of Pytorch and resolved the issue ？

May 16 '23 10:05 JaniceSu-su

same problem while using torch==1.9.1+cu111, and ./make.sh doesnt work as well

Hi, please install PyTorch>=1.10.0.

Thank you so much for your quick reply, I have realised this problem and I'm installing PyTorch==1.10.0 Thanks again : )

same problem while using torch==1.9.1+cu111, and ./make.sh doesnt work as well

Hi, please install PyTorch>=1.10.0.

Thank you so much for your quick reply, I have realised this problem and I'm installing PyTorch==1.10.0 Thanks again : )

Hello, have you installed a higher version of Pytorch and resolved the issue ？

hello! After change torch==1.11.0+cu113 to 1.13.1+cu117, I encountered the same error. Have you resolved the issue😟

Sep 06 '23 15:09 H-Ambrose