AssertionError with latest spconv version

[Jamie725]

As title. This happens at spconv/spconv/utils/init.py.
Current version of spconv of this file has this default setting:

def points_to_voxel( ...
full_mean=False,
block_filtering=True,
...):

if full_mean:
assert block_filtering is False

This will result in assertion error.

If I directly change block_filtering=False, there will be other issues.
Is there any suggestions for this version of spconv?
Could you please help me with this issue? Thanks!!!

[pangsu0613]

Hi @Jamie725 , you need to use the spconv with the commit that I mentioned in the readme. This is constraint by SECOND-V1.5.

[Jamie725]

Hi @pangsu0613,
Thanks for the information. I followed the following commands but failed building spconv:

git clone https://github.com/traveller59/spconv.git
cd spconv
git checkout 8da6f967fb9a054d8870c3515b1b44eca2103634

I got building errors with:
error: no matching function for call to ‘torch::jit::RegisterOperators::RegisterOperators(const char [28], )’
20 | torch::jit::RegisterOperators("spconv::get_indice_pairs_2d", &spconv::getIndicePair<2>)

Therefore, I was wondering if I did anything wrong. Could you please share with me the some suggested instructions to install this specific commit on spconv?

Thank you very much for your help!

[pangsu0613]

Hello @Jamie725 , I understand installing spconv is not easy. You can try change torch::jit::RegisterOperators into torch::RegisterOperators() under /spconv/src/spconv/all.cc.
You can also check the issues section in spconv repo for other instructions.

[Jamie725]

Hi @pangsu0613,
It works! Thank you very much for your help!

[Anirbanbhk88]

@Jamie725 how did you solve the above issue