Import Error for spconv-cu117

[oneya999]

Hi SpConv author, thanks for your excellent work! When I train my model but there is always an import error after installing spconv :

raceback (most recent call last):
  File "/home/user/IoSI-CP/IoSI-CP/opencood/data_utils/pre_processor/sp_voxel_preprocessor.py", line 24, in __init__
    from spconv.utils import VoxelGeneratorV2 as VoxelGenerator
  File "/home/user/anaconda3/envs/IoSI-CP/lib/python3.8/site-packages/spconv/utils/__init__.py", line 18, in <module>
    from spconv.cppconstants import CPU_ONLY_BUILD
  File "/home/user/anaconda3/envs/IoSI-CP/lib/python3.8/site-packages/spconv/cppconstants.py", line 15, in <module>
    import spconv.core_cc as _ext
ImportError: arg(): could not convert default argument 'workspace: tv::Tensor' in method '<class 'spconv.core_cc.csrc.sparse.convops.gemmops.GemmTunerSimple'>.run_with_tuned_result' into a Python object (type not registered yet?)

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "opencood/tools/train.py", line 209, in <module>
    main()
  File "opencood/tools/train.py", line 45, in main
    opencood_train_dataset = build_dataset(hypes, visualize=False, train=True,uni_time_delay=-1)
  File "/home/user/IoSI-CP/IoSI-CP/opencood/data_utils/datasets/__init__.py", line 31, in build_dataset
    dataset = __all__[dataset_name](
  File "/home/user/IoSI-CP/IoSI-CP/opencood/data_utils/datasets/intermediate_fusion_dataset.py", line 52, in __init__
    self.pre_processor = build_preprocessor(params['preprocess'],
  File "/home/user/IoSI-CP/IoSI-CP/opencood/data_utils/pre_processor/__init__.py", line 27, in build_preprocessor
    processor = __all__[process_method_name](
  File "/home/user/IoSI-CP/IoSI-CP/opencood/data_utils/pre_processor/sp_voxel_preprocessor.py", line 27, in __init__
    from spconv.utils import Point2VoxelCPU3d as VoxelGenerator
  File "/home/user/anaconda3/envs/IoSI-CP/lib/python3.8/site-packages/spconv/utils/__init__.py", line 18, in <module>
    from spconv.cppconstants import CPU_ONLY_BUILD
  File "/home/user/anaconda3/envs/IoSI-CP/lib/python3.8/site-packages/spconv/cppconstants.py", line 15, in <module>
    import spconv.core_cc as _ext
ImportError: generic_type: cannot initialize type "ExternalAllocator": an object with that name is already defined

My spconv vresion is spconv-cu117 2.3.6 , torch vrsion is 1.13.0+cu117, I look forward to any suggestions.

[kanglang123]

Me too.

[vehxianfish]

Based on ReadME. The reason is most likely by cumm and spconv installation question. My solution is to uninstall cumm, spconv and reinstall it.

1. uninstall spconv

   pip list | grep spconv 
   pip uninstall all-spconv -relative

   pip list | grep cumm
   pip uninstall all-cumm-relative

2. find match version with your cuda

   pip install spconv-cu{version}

3. if still not ok, you should try to downgrade spconv version. because in ReadME, it's safe to have different minor cuda version between system and conda (pytorch) in CUDA >= 11.0 because of CUDA Minor Version Compatibility. For example, you can use spconv-cu114 with anaconda version of pytorch cuda 11.1 in a OS with CUDA 11.2 installed.

[colorfulandcjy0806]

Perhaps everyone can try this. My environment is ubuntu20.04 and cuda is 11.4. Although I tried many methods but found them ineffective, I successfully solved the above problem through pip install spconv-cu102. I hope it is helpful to everyone.😄