yukang

Hi, I think this issue is from the version issue of pccm and setuptools. Would you please provide me a detailed environment list, via `conda list`? Regards, Yukang Chen

Hi, Would you please have a try with setuptools==58.0.4? Regards, Yukang Chen

Hi, This is weird. I think after you compiled. What you need to do is to install the complied file. `pip install path_to_the_spconv_wheel.whl`. Instead of `import spconv` directly. Anyway, I...

Sure! [environment.yml.zip](https://github.com/dvlab-research/spconv-plus/files/11317599/environment.yml.zip)

Sorry for the late reply. Would you please provide the version information, torch and spconv?

Hi, the number of groups is the numel of the splits. For example, "spatial-wise 7×7×7 convolutions is grouped into 3×3×3 splits" is 3x3x3 = 27

Sorry for the late reply. Would you please provide the version information, torch and spconv?

Would you please have a try on spconv 2.1.x?

Hi @fjzpcmj ， Thanks for your interests in our work. Sorry for the late reply. I have some deadline this week. I will check the nusc_centerpoint_voxelnet_0075voxel_fix_bn_z_largekernel3d_large.py. I used 4 GPUs...

Thanks for your message. Generally, "large" performs a bit better than "tiny" (less than 0.5 mAP). But "tiny" is more stable and faster.