search

zzangjinsun / NLSPN_ECCV20

Park et al., Non-Local Spatial Propagation Network for Depth Completion, ECCV, 2020
MIT License
321 stars 57 forks source link

Work with Cuda 11 #44

Closed s-mostafa-a closed 2 years ago

s-mostafa-a commented 2 years ago

Hi I hope you're doing great. Is there any solution to work with later version of CUDA and its related torch, apex, etc? I tried to install apex but it seems that it's not working for my gpu.

zzangjinsun commented 2 years ago

Hi @s-mostafa-a , I am using apex with python 3.8, cuda-11.1, apex 0.1, and pytorch 1.9.1 now and there is no problem. Which problem do you have?

s-mostafa-a commented 2 years ago

Thank you for your reply @zzangjinsun. I installed the versions you mentioned successfully. But still I have problem installing DCN.

s-mostafa-a commented 2 years ago

Never mind. I probably found the solution. There is some problem here with the modified repository of dcn. They have fixed it here in "pytorch_1.0.0" branch of the original repository of dcn. (Mainly this commit) What I did is:

git clone https://github.com/chengdazhi/Deformable-Convolution-V2-PyTorch/
cd ./Deformable-Convolution-V2-PyTorch/
git checkout pytorch_1.0.0
sh make.sh
zura-false commented 2 years ago

Hi, I am also using these configs but I am getting an error when installing apex，I got the error" fatal: Could not parse object '4ef930c1c884fdca5f472ab2ce7cb9b505d26c1a' "by running"git reset --hard 4ef930c1c884fdca5f472ab2ce7cb9b505d26c1a",If I run the next sentence of pip directly，It will prompt the error of the cuda version, have you encountered this situation?

zura-false commented 2 years ago

Hi, I am also using these configs but I am getting an error when installing apex，I got the error" fatal: Could not parse object '4ef930c1c884fdca5f472ab2ce7cb9b505d26c1a' "by running"git reset --hard 4ef930c1c884fdca5f472ab2ce7cb9b505d26c1a",If I run the next sentence of pip directly，It will prompt the error of the cuda version, have you encountered this situation?

I replaced the correct torch version, it can run after skipping this sentence