fudan-zvg / DeepInteraction

[NeurIPS 2022] DeepInteraction: 3D Object Detection via Modality Interaction
MIT License
221 stars 16 forks source link

Conflict from the environment. #11

Closed jarvishou829 closed 1 year ago

jarvishou829 commented 1 year ago

Thanks for your great work. But when I prepare the environment following the install.md error happens. First, the title mismatch with the command in theinstall.md to install mmcv-full. One is 1.4.1 and the other 1.3.18. What's more, the numpy and ypy-websocket version conflict when installing mmdet3d 0.17.1. After I manage these problems, when I run the train command, the error goes behind. It seems the mismatch of pytorch and mmcv version according to . Hope you can provide some advice. Thanks.

image

Alexander0Yang commented 1 year ago

One is 1.4.1 and the other 1.3.18.

My typo. The version of mmcv-full should be 1.3.18. I have corrected it in the install.md.

What's more, the numpy and ypy-websocket version conflict when installing mmdet3d 0.17.1.

I remember that the mmdet3d 0.17.1 requires numpy < 1.20.0. If the numpy installed in the previous steps is higher than this, you can manually run pip install -r requirements/runtime.txt to obtain the dependency with compatible version.

After I manage these problems, when I run the train command, the error goes behind.

Seems the mmcv is not installed successfully. Can you run python mmdet3d/utils/collect_env.py to collect necessary environment information and paste it here?

jarvishou829 commented 1 year ago

One is 1.4.1 and the other 1.3.18.

My typo. The version of mmcv-full should be 1.3.18. I have corrected it in the install.md.

What's more, the numpy and ypy-websocket version conflict when installing mmdet3d 0.17.1.

I remember that the mmdet3d 0.17.1 requires numpy < 1.20.0. If the numpy installed in the previous steps is higher than this, you can manually run pip install -r requirements/runtime.txt to obtain the dependency with compatible version.

After I manage these problems, when I run the train command, the error goes behind.

Seems the mmcv is not installed successfully. Can you run python mmdet3d/utils/collect_env.py to collect necessary environment information and paste it here?

The collect_env.py reports the same error. By the way, I'm not sure if there's a problem with the python version. Your suggestion is 3.7, but when I install mmde3d 0.17.1, it requires 3.8 at least so I upgrade it. May be I have to try re-create a conda environment. image

Alexander0Yang commented 1 year ago

when I install mmde3d 0.17.1, it requires 3.8 at least so I upgrade it.

As far as I know, mmdet3d 0.17.1 is compatible with python 3.7.

After I manage these problems,

May I ask how you manage these problems. It does seem that the mmcv was not built correctly.

Anyway, you can feel at ease to install the environment using the recommended version of each library.

jarvishou829 commented 1 year ago

python mmdet3d/utils/collect_env.py

Thanks for your kind reply. I create a new environment and strictly follows the install.md, but errors still appear. As you can see, my env is below. env I follow your commands to install mmcv 1.3.18, mmdet 2.14.0, mmseg 0.14.1. But when I compile python setup.py install to install mmdet3d 0.17.1. The error appears error1 It seems the problem of scipy version. So I try to run pip install -r requirements.ttxt. After that I compile python setup.py install, no error appears. However, when I try to train, The same error goes. final_error Would you please double check the mmlab packages version in your environment?

loongofqiao commented 1 year ago

python mmdet3d/utils/collect_env.py

Thanks for your kind reply. I create a new environment and strictly follows the install.md, but errors still appear. As you can see, my env is below. env I follow your commands to install mmcv 1.3.18, mmdet 2.14.0, mmseg 0.14.1. But when I compile python setup.py install to install mmdet3d 0.17.1. The error appears error1 It seems the problem of scipy version. So I try to run pip install -r requirements.ttxt. After that I compile python setup.py install, no error appears. However, when I try to train, The same error goes. final_error Would you please double check the mmlab packages version in your environment?

Same error,i also strictly follows the install.md

Alexander0Yang commented 1 year ago

Here is the environment we used.

TorchVision: 0.10.1+cu111
OpenCV: 4.5.5
MMCV: 1.3.18
MMCV Compiler: GCC 7.3
MMCV CUDA Compiler: 11.1
MMDetection: 2.14.0
MMSegmentation: 0.14.1
MMDetection3D: 0.17.1+f110797

This problem may be because the inconsistence between the environments where mmcv is built by and running on, as is suggested in the faq doc.

Can you run python -c 'import mmcv; import mmcv.ops' to check whether your mmcv is correctly installed?

jarvishou829 commented 1 year ago

Here is the environment we used.

TorchVision: 0.10.1+cu111
OpenCV: 4.5.5
MMCV: 1.3.18
MMCV Compiler: GCC 7.3
MMCV CUDA Compiler: 11.1
MMDetection: 2.14.0
MMSegmentation: 0.14.1
MMDetection3D: 0.17.1+f110797

This problem may be because the inconsistence between the environments where mmcv is built by and running on, as is suggested in the faq doc.

Can you run python -c 'import mmcv; import mmcv.ops' to check whether mmcv is correctly installed?

If the same error occurs again, you can try to recompile mmcv-full by running pip install mmcv-full==1.3.18 -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.9.0/index.html

I think I address this problem. I solved this by installing mmcv-full, 'mmdet', 'mmsegmentation' using the officially suggested openmim instead of pip. Nothing else is changed, so I think it may be the pip installation problem, and I suggest the mim commands. @loongofqiao

pip install -U openmim
mim install mmcv-full==1.3.18
mim install mmdet==2.14.0
mim install mmsegmentation==0.14.1

Now my env is

TorchVision: 0.10.1+cu111
OpenCV: 4.6.0
MMCV: 1.3.18
MMCV Compiler: GCC 7.3
MMCV CUDA Compiler: 11.1
MMDetection: 2.14.0
MMSegmentation: 0.14.1
MMDetection3D: 0.17.1+
loongofqiao commented 1 year ago

Here is the environment we used.

TorchVision: 0.10.1+cu111
OpenCV: 4.5.5
MMCV: 1.3.18
MMCV Compiler: GCC 7.3
MMCV CUDA Compiler: 11.1
MMDetection: 2.14.0
MMSegmentation: 0.14.1
MMDetection3D: 0.17.1+f110797

This problem may be because the inconsistence between the environments where mmcv is built by and running on, as is suggested in the faq doc. Can you run python -c 'import mmcv; import mmcv.ops' to check whether mmcv is correctly installed? If the same error occurs again, you can try to recompile mmcv-full by running pip install mmcv-full==1.3.18 -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.9.0/index.html

I think I address this problem. I solved this by installing mmcv-full, 'mmdet', 'mmsegmentation' using the officially suggested openmim instead of pip. Nothing else is changed, so I think it may be the pip installation problem, and I suggest the mim commands. @loongofqiao

pip install -U openmim
mim install mmcv-full==1.3.18
mim install mmdet==2.14.0
mim install mmsegmentation==0.14.1

Now my env is

TorchVision: 0.10.1+cu111
OpenCV: 4.6.0
MMCV: 1.3.18
MMCV Compiler: GCC 7.3
MMCV CUDA Compiler: 11.1
MMDetection: 2.14.0
MMSegmentation: 0.14.1
MMDetection3D: 0.17.1+

ok,Thank you

sunnyHelen commented 1 year ago

Here is the environment we used.

TorchVision: 0.10.1+cu111
OpenCV: 4.5.5
MMCV: 1.3.18
MMCV Compiler: GCC 7.3
MMCV CUDA Compiler: 11.1
MMDetection: 2.14.0
MMSegmentation: 0.14.1
MMDetection3D: 0.17.1+f110797

This problem may be because the inconsistence between the environments where mmcv is built by and running on, as is suggested in the faq doc. Can you run python -c 'import mmcv; import mmcv.ops' to check whether mmcv is correctly installed? If the same error occurs again, you can try to recompile mmcv-full by running pip install mmcv-full==1.3.18 -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.9.0/index.html

I think I address this problem. I solved this by installing mmcv-full, 'mmdet', 'mmsegmentation' using the officially suggested openmim instead of pip. Nothing else is changed, so I think it may be the pip installation problem, and I suggest the mim commands. @loongofqiao

pip install -U openmim
mim install mmcv-full==1.3.18
mim install mmdet==2.14.0
mim install mmsegmentation==0.14.1

Now my env is

TorchVision: 0.10.1+cu111
OpenCV: 4.6.0
MMCV: 1.3.18
MMCV Compiler: GCC 7.3
MMCV CUDA Compiler: 11.1
MMDetection: 2.14.0
MMSegmentation: 0.14.1
MMDetection3D: 0.17.1+

I also encounter the same error when using "python setup.py install".

image

I tried the mim install ... way, but still got the same error. Is there any suggestion to fix it?

Alexander0Yang commented 1 year ago

Hi, @sunnyHelen You can run pip install -r requirements.txt before python setup.py install. Please refer to @jarvishou829‘s solution.

It seems the problem of scipy version. So I try to run pip install -r requirements.ttxt. After that I compile python setup.py install, no error appears.

sunnyHelen commented 1 year ago

Thanks for your reply. Now the mmdet3d seems built successfully, but there's another error.

image
Alexander0Yang commented 1 year ago

That’s because your mmcv was not installed successfully. You can try to reinstall it by running pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu111/torch1.10.0/index.html.

sunnyHelen commented 1 year ago

Yeah, problem fixed. I didn't work because I didn't uninstall the wrong version. Now the environment issue is addressed. Thank you very much.