ImportError: DLL load failed while importing _ext: The specified procedure could not be found.

[gitgitgithut]

Hi there, I recently installed mmcv-full 1.4.1 and mmdet 2.19.1 using conda and openmim on my windows 10 device, and encountered an import error when running the following line:
from mmdet.apis import init_detector, inference_detector, show_result_pyplot
The error messages I got were:

I double checked the versions of my cudatoolkit, torch, mmcv-full and mmdet, and everything seems fine to me. I have pytorch==1.10.0, cudatoolkit==11.3.1, mmcv-full==1.4.1 and mmdet==2.19.1.

I tried to reinstall mmcv-full to check I got the build version correctly, here's the result:

Yet the same error messages were printed. I suspect it is still an compatibility issue between mmcv-full and mmdet, but i think I got the installation steps correct. Would you please help me with this issue? Thank you.

The following are all the packages I have installed in my venv:

[zhouzaida]

Hi, did you install mmcv (not mmcv-full) before?

Note: Do not install both versions in the same environment, otherwise you may encounter errors like ModuleNotFound.

[gitgitgithut]

Hi, did you install mmcv (not mmcv-full) before?

Note: Do not install both versions in the same environment, otherwise you may encounter errors like ModuleNotFound.

No, I have re-created my venv and I am certain that only mmcv-full is installed.

[del-zhenwu]

Hi, did you install mmcv (not mmcv-full) before?
Note: Do not install both versions in the same environment, otherwise you may encounter errors like ModuleNotFound.

No, I have re-created my venv and I am certain that only mmcv-full is installed.

I have uploaded the zip file for mmdet2.19.1-mmcv1.4.1-torch1.10.0-py38，would you give a try with installing the whl file?
mmdet-2.19.1-py3-none-any.whl.zip

[del-zhenwu]

@gitgitgithut This package has been proved OK in my env.

[gitgitgithut]

@gitgitgithut This package has been proved OK in my env.

Hi there, I re-installed my venv using the .whl file you provided me, and the problem persists.
Here are my installation steps:

1. create a new conda venv with python==3.8.12
2. use pip to install mmdet from the wheel you provided me:
   pip install mmdet-2.19.1-py3-none-any.whl
3. use conda to install pytorch and cudatoolkits with:
   conda install pytorch==1.10.0 torchvision torchaudio cudatoolkit=11.3 -c pytorch
4. use pip to install opencv-python-headless and openmim with:
   pip install opencv-python-headless openmim
5. use openmim to install mmcv-full:
   mim install mmcv-full

Here's the output of running the same code:

The following is the installed packages' info:

[gitgitgithut]

I just found out there are also some issues with mmcv:

[zhouzaida]

Hi @gitgitgithut , we have updated the pre-built packages of 1.4.1. Please have a try.

[gitgitgithut]

Hi @zhouzaida @del-zhenwu, I have reinstalled my venv with the updated mmcv-full 1.4.2, the issue is now resolved:

Really appreciate your efforts, Thank you!

[YellowMatthew]

That helps much!!

[duduzai2019]

I have reinstalled my venv with the updated mmcv-full 1.4.2, the issue is now resolved

Hi @zhouzaida @del-zhenwu, I have reinstalled my venv with the updated mmcv-full 1.4.2, the issue is now resolved: Really appreciate your efforts, Thank you!

hi, How can installI mmcv-full 1.4.2 ??
Thanks!

[gitgitgithut]

Hi @duduzai2019, as the official mmcv documentation stated, you can use pip to install a specific version of mmcv-full with pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html.

[duduzai2019]

嗨，@duduzai2019，正如官方mmcv文档所述，您可以使用pip安装特定版本的mmcv-full。pip install mmcv-full=={mmcv_version} -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html

Thank you for your answer, It has been replaced by mvcc 1.4.2 , but train.py is still an old proble

[aldazhu]

hi

Encountered the same error in mmcv 2.0.0rc4 when attempting to install it using "pip install mmcv". However, I was able to resolve the issue by building mmcv from the source code。
gpu : Quadro M4000
cuda: 11.6
CUDA Capability: 5.2
windows10
The usage of 'build' can be referred to in this link：https://mmcv.readthedocs.io/en/latest/get_started/build.html

[lsewcx]

已解决
原因可能使用pip安装
解决方法改用mim安装就成功了

[cybbbbbbb]

hi @gitgitgithut. I have the same problem as you, and it has been bothering me for three days. But I finally solved it now.

[vigneshwar1565]

Hi @cybbbbbbb , I'm facing the same issue.
I have installed mmcv - 2.0.0
when I do python demo/demo_skeleton.py demo.mp4 short_serve_out.mp4 --config slowonly_r50_8xb16-u48-240e_ntu60-xsub-keypoint,py --checkpoint slowonly_r50_8xb16-u48-240e_ntu60-xsub-keypoint_20220815-38db104b.pth --label-map tools/data/skeleton/label_map_ntu60.txt
Im getting ImportError: DLL load failed while importing _ext: The specified procedure could not be found.

[vigneshwar1565]

Hi @cybbbbbbb , I'm facing the same issue. I have installed mmcv - 2.0.0 when I do python demo/demo_skeleton.py demo.mp4 short_serve_out.mp4 --config slowonly_r50_8xb16-u48-240e_ntu60-xsub-keypoint,py --checkpoint slowonly_r50_8xb16-u48-240e_ntu60-xsub-keypoint_20220815-38db104b.pth --label-map tools/data/skeleton/label_map_ntu60.txt Im getting ImportError: DLL load failed while importing _ext: The specified procedure could not be found.

I was able to fix it by downgrading my pytorch to 2.0.0
pip install torch==2.0.0 torchvision==0.15.1 torchaudio==2.0.1 --index-url https://download.pytorch.org/whl/cu118