megvii-research / CVPR2023-UniDistill

CVPR2023 (highlight) - UniDistill: A Universal Cross-Modality Knowledge Distillation Framework for 3D Object Detection in Bird's-Eye View
Apache License 2.0
101 stars 10 forks source link

Error when import `voxel_pooling_ext` package #3

Closed SeaBird-Go closed 1 year ago

SeaBird-Go commented 1 year ago

Hi, thanks for sharing this code.

I found you just upload the voxel_pooling_ext.cpython-36m-x86_64-linux-gnu.so file for this voxel_pooling_ext module. However when I ran the code, I got the error:

ImportError: ~/CVPR2023-UniDistill/unidistill/layers/blocks_3d/mmdet3d/voxel_pooling_ext.cpython-36m-x86_64-linux-gnu.so: undefined symbol: _ZNK2at6Tensor8data_ptrIfEEPT_v

Could you help me solve this problem or share the source code about this module?

FudongGe commented 1 year ago

Good work! But I met the same 'import' question. And could you introduce how you get the file 'voxel_pooling_ext.cpython-36m-x86_64-linux-gnu.so'? Thanks very much!

kebijuelun commented 1 year ago

Hi, thank you for your attention to our work. @FudongGe @SeaBird-1995

This issue is likely caused by a mismatch in the Python version. The version we currently need is as follows:

Install python3.6 + CUDAv10.2 + CUDNNv8.0.4 + pytorch(v1.9.0)

pip3 install mmcv-full==1.4.2 -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.9.0/index.html
pip3 install mmdet==2.20.0
# download mmdet3d whl from https://drive.google.com/file/d/1y6AjikFQGc400dTim9IIFdvR8RNvnPId/view?usp=share_link
pip3 install mmdet3d-0.18.0-cp36-cp36m-linux_x86_64.whl

The most convenient reproduction method currently is to use the Conda environment to ensure that the Python version is consistent with what we described. (for emample: conda create -n unidistill python==3.6; conda activate unidistill)

Sorry for any inconvenience. And we will confirm whether the relevant source code here can be opened source recently, and if so, it can reduce the version restrictions.

kebijuelun commented 1 year ago

Hi, thank you for your attention to our work. @FudongGe @SeaBird-1995

This issue is likely caused by a mismatch in the Python version. The version we currently need is as follows:

Install python3.6 + CUDAv10.2 + CUDNNv8.0.4 + pytorch(v1.9.0)

pip3 install mmcv-full==1.4.2 -f https://download.openmmlab.com/mmcv/dist/cu102/torch1.9.0/index.html
pip3 install mmdet==2.20.0
# download mmdet3d whl from https://drive.google.com/file/d/1y6AjikFQGc400dTim9IIFdvR8RNvnPId/view?usp=share_link
pip3 install mmdet3d-0.18.0-cp36-cp36m-linux_x86_64.whl

The most convenient reproduction method currently is to use the Conda environment to ensure that the Python version is consistent with what we described. (for emample: conda create -n unidistill python==3.6; conda activate unidistill)

Sorry for any inconvenience. And we will confirm whether the relevant source code here can be opened source recently, and if so, it can reduce the version restrictions.

SeaBird-Go commented 1 year ago

Got that. Thanks for your reply. Hopy for openining the original sources to ease the environment constraints.