mit-han-lab / spvnas

[ECCV 2020] Searching Efficient 3D Architectures with Sparse Point-Voxel Convolution
http://spvnas.mit.edu/
MIT License
587 stars 109 forks source link

CUDA error: invalid configuration argument #92

Closed tomanick closed 2 years ago

tomanick commented 2 years ago

When I wanted to run train.py file, I got this error.

cuda_error

Does anybody know how to solve this problem?

Environment: python: 3.7 pytorch: 1.10.1 Cuda: 11.3 Cudnn: 8.2.1

CCInc commented 2 years ago

Hi @tomanick , what GPU / compute capability are you trying to run on?

tomanick commented 2 years ago

@CCInc , NVIDIA Geforce RTX 3080Ti

tomanick commented 2 years ago

I guessed that perhaps this problem had something to do with OOM. Afterwards, I found that my input data format of point clouds was [xyz, pc numbers] which was the input format of pvcnn. The point clouds format for sparse_quantize of spvcnn is [pc numbers, xyz]. When I changed the order of point cloud data, the problem was solved. Thanks!