RunError

[JohnnyKimura]

您好，我在pip 安装submodels中的三个对应模块时都出现了如下问题。我尝试修改我自己的cuda版本到11.6，发现还是出现这个问题，请问有什么解决办法吗，非常感谢！

[kcheng1021]

You could follow the environment setup in 3DGS first. Then adding the additional package in our project. It seems there are mismatching in CUDA version, (maybe the conflict is from /usr/local/cuda in .bashrc and the pytorch, cudatoolkit in your conda env? I am not sure about this).

But at least our modification has no conflict with the original 3DGS. So maybe you can make sure 3DGS is loaded successfully, such as the diff_gaussian_rasterization module can be installed correctly.

[JohnnyKimura]

hi there,i really appreciate your amazing work. according to my problem，i have tried to change the version of CUDA from 12.0 to 11.6，but it encounter the same question. i ran the code on our sever station and there are no other authority for me to do more things，so change the cuda version is the least i can do according to the environment.yml . so i start to think maybe the problem comes from environment setup. i try to create virtual environment on my sever station by using the pytorch and cuda toolkit that satisfy my CUDA 12.0，it did work when i pip" diff-gaussian "and "simple-knn"，but it still encount the same problem like the pics up there when i tried to pip "propagation" module. i have tried to fix the problem a whole afternoon tody,but still got no clues. sorry again to bother u again，i would be so grateful to your reply

[kcheng1021]

Sorry to hear that. Based on your CUDA 12.0, have you changed the 'include_dir' in setup.py of Propagation to adapt to your CUDA path?

[JohnnyKimura]

Hi there，i did modify it ，but dont know whether it can recognize my path or not. 😂，but when i modify it and ran the code it still encounter the same problem.. any suggestion？ maybe i will tried again tomorrow😫😰

[kcheng1021]

What is your pytorch version? The error seems saying the cuda mismatching from include path and the version compile pytorch.

[JohnnyKimura]

i tried the environment.yml and downgrade the version of my CUDA from 12.0 to 11.6 ，encounter the problem above，maybe my installation of cuda 11.6 is wrong . But i also tried to download pytorch that support cuda 12.0（according to https://pytorch.org/） it still remain the same ，a little bit confused and frustrated today ， but i will try again to figure it out

[JohnnyKimura]

Hi there，i strictly follow your environment.yaml and downgrade my cuda to 11.7，i successfully install the submodules，but when i run the bash.sh (waymo scene)，i encounter a problem： ImportError: /home/stlyj/.conda/envs/gaussianpro/lib/python3.7/site-packages/gaussianpro.cpython-37m-x86_64-linux-gnu.so: undefined symbol: _ZN2cv5errorEiRKSsPKcS3_i

i search the solution but find little helpful answer， they cant solve my problem for now：（i have tried opencv ver from 4.2 to 4.5.5，i even tried 3.9.5，same problem） https://forum.opencv.org/t/undefined-symbol-zn2cv5erroreirksspkcs3-i/8050

any suggestion？thanks for your kind reply @kcheng1021

[kcheng1021]

It seems that your machine is not dynamically linked to the OpenCV library or there is a version mismatch, resulting in the missing CV-related dependencies. You can try linking to the OpenCV library in the Conda environment (conda install -c conda-forge opencv) instead of the system-installed OpenCV library obtained through apt install.