No module named 'pointops_cuda'

[onikiri17]

I've attempted to install PyTorch 1.12 with CUDA 11.6, along with several alternative versions, and recompiled. Additionally, I've tried to compile the setup.py file independently, but I continue to face the error described in the title. Would anyone happen to have encountered a similar situation? Could you kindly share how it might be resolved? Thank you in advance for any assistance.

[CuiRuikai]

Could you please provide your error information?
The pointops lib was adopted from Point Transformer. It is indeed difficult to compile.
You can possibly try:
POSTECH-CVLab/point-transformer#56
POSTECH-CVLab/point-transformer#82
POSTECH-CVLab/point-transformer#73
etc.

[xingjinderhuoche]

I also met "ModuleNotFoundError: No module named 'pointops_cuda' "problem. And I only need "pointops_cuda.furthestsampling_cuda(b, n_max, xyz, offset, new_offset, tmp, idx)" and " pointops_cuda.knnquery_cuda(m, nsample, xyz, new_xyz, offset, new_offset, idx, dist2)" function。Could you please provide the code of these two functions。Thanks a lot！

[xingjinderhuoche]

I found that these two functions are quite complicated, is c++, I look at other methods.

[dperozzi]

This issue is solved by doing the following

cd extensions/pointops
FORCE_CUDA=1 python setup.py install

Hope this helps!

[xingjinderhuoche]

Thanks!

This issue is solved by doing the following

cd extensions/pointops
FORCE_CUDA=1 python setup.py install

Hope this helps!