ImportError: No module named 'voxelize_cuda'

[jiangsijia9495]

I am keep getting no module named 'voxelize_cuda' error when run python main_test.py.
I am using pytorch 1.7.0
Cuda 10.1
Could anyone tell me where could possible be wrong? Thx

Using /home/sijia/.cache/torch_extensions as PyTorch extensions root...
Detected CUDA files, patching ldflags
Emitting ninja build file /home/sijia/.cache/torch_extensions/voxelize_cuda/build.ninja...
Building extension module voxelize_cuda...
Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)
1.10.2
Loading extension module voxelize_cuda...
Traceback (most recent call last):
File "main_test.py", line 133, in
main_test_wo_gt_smpl_with_optm(input_image_dir,
File "main_test.py", line 36, in main_test_wo_gt_smpl_with_optm
from evaluator import Evaluator
File "/home/sijia/Desktop/PaMIR/networks/evaluator.py", line 23, in
from neural_voxelization_layer.voxelize import Voxelization
File "/home/sijia/Desktop/PaMIR/networks/neural_voxelization_layer/voxelize.py", line 13, in
voxelize_cuda = load(
File "/home/sijia/anaconda3/envs/pamir/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 969, in load
return _jit_compile(
File "/home/sijia/anaconda3/envs/pamir/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 1196, in _jit_compile
return _import_module_from_library(name, build_directory, is_python_module)
File "/home/sijia/anaconda3/envs/pamir/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 1543, in _import_module_from_library
file, path, description = imp.find_module(module_name, [path])
File "/home/sijia/anaconda3/envs/pamir/lib/python3.8/imp.py", line 296, in find_module
raise ImportError(_ERR_MSG.format(name), name=name)
ImportError: No module named 'voxelize_cuda'
(pamir) sijia@sijia-750-170se:~/Desktop/PaMIR/networks$ python3 main_test.py
True
Using /home/sijia/.cache/torch_extensions as PyTorch extensions root...
Detected CUDA files, patching ldflags
Emitting ninja build file /home/sijia/.cache/torch_extensions/voxelize_cuda/build.ninja...
Building extension module voxelize_cuda...
Allowing ninja to set a default number of workers... (overridable by setting the environment variable MAX_JOBS=N)
1.10.2
Loading extension module voxelize_cuda...
Traceback (most recent call last):
File "main_test.py", line 133, in
main_test_wo_gt_smpl_with_optm(input_image_dir,
File "main_test.py", line 36, in main_test_wo_gt_smpl_with_optm
from evaluator import Evaluator
File "/home/sijia/Desktop/PaMIR/networks/evaluator.py", line 23, in
from neural_voxelization_layer.voxelize import Voxelization
File "/home/sijia/Desktop/PaMIR/networks/neural_voxelization_layer/voxelize.py", line 13, in
voxelize_cuda = load(
File "/home/sijia/anaconda3/envs/pamir/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 969, in load
return _jit_compile(
File "/home/sijia/anaconda3/envs/pamir/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 1196, in _jit_compile
return _import_module_from_library(name, build_directory, is_python_module)
File "/home/sijia/anaconda3/envs/pamir/lib/python3.8/site-packages/torch/utils/cpp_extension.py", line 1543, in _import_module_from_library
file, path, description = imp.find_module(module_name, [path])
File "/home/sijia/anaconda3/envs/pamir/lib/python3.8/imp.py", line 296, in find_module
raise ImportError(_ERR_MSG.format(name), name=name)
ImportError: No module named 'voxelize_cuda'

[jiangsijia9495]

Fixed.
Cause of the issue: mismatch of the gcc, cuda versions (pytorch pre-compiled version made things worse).

Ubuntu: 18.04

Fix:

Install cuda 10.0 (https://developer.nvidia.com/cuda-10.0-download-archive)
Switch to gcc 7.3 (if you have many versions of gcc use update-alternatives. Ubuntu 18.04 comes by default with gcc 7.3.)
Install the pytoch pip3 install torch==1.7.0 torchvision==0.8.1 -f https://download.pytorch.org/whl/cu101/torch_stable.html
If you have many versions of cuda: add the following to ~/.bashrc and specify the cuda version you intend to use:
CUDAVER=cuda-10.0
export PATH=/usr/local/$CUDAVER/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/$CUDAVER/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=/usr/local/$CUDAVER/lib64:$LD_LIBRARY_PATH
export CUDA_PATH=/usr/local/$CUDAVER
export CUDA_ROOT=/usr/local/$CUDAVER
export CUDA_HOME=/usr/local/$CUDAVER
export CUDA_HOST_COMPILER=/usr/bin/gcc-7.3
Closed.

[jiangsijia9495]

add some tips about trimesh. If you find mesh.contains() return incorrect points. Check this link mikedh/trimesh#242 (comment)

[zoroinparis]

Fixed.
Cause of the issue: mismatch of the gcc, cuda versions (pytorch pre-compiled version made things worse).

Ubuntu: 18.04

Fix:

Install cuda 10.0 (https://developer.nvidia.com/cuda-10.0-download-archive)
Switch to gcc 7.3 (if you have many versions of gcc use update-alternatives. Ubuntu 18.04 comes by default with gcc 7.3.)
Install the pytoch pip3 install torch==1.7.0 torchvision==0.8.1 -f https://download.pytorch.org/whl/cu101/torch_stable.html
If you have many versions of cuda: add the following to ~/.bashrc and specify the cuda version you intend to use:
CUDAVER=cuda-10.0
export PATH=/usr/local/$CUDAVER/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/$CUDAVER/lib:$LD_LIBRARY_PATH
export LD_LIBRARY_PATH=/usr/local/$CUDAVER/lib64:$LD_LIBRARY_PATH
export CUDA_PATH=/usr/local/$CUDAVER
export CUDA_ROOT=/usr/local/$CUDAVER
export CUDA_HOME=/usr/local/$CUDAVER
export CUDA_HOST_COMPILER=/usr/bin/gcc-7.3
Closed.

[zoroinparis]

I got the same problem ,and i switch the version but nothing worked .Do you have any idea?

[jiangsijia9495]

FYI.
Evaluation CrisHY1995/StereoPIFu_Code#1 (comment)
And PIFuHD lib/evaluator

vertice.txt
Local class used to load and store SMPL's vertices coordinate at rest pose at mean shape

[jiangsijia9495]

installwifi.txt
For RTX 3090
Cuda 11.1, V11.1.105
Driver Version: 470.57.02