spconv vs torchsparse

[SandeepMenonKodiak]

Hi
Thank you for consolidating SPVCNN and MinkowskiNet along with your work.
I wanted to ask if you tried implementing your model using torchsparse. Any reason for choosing spconv (featurewise or performance wise).
In your codebase I see SparseSequential, SparseConvTensor, SubMConv3d from spconv. They can be substituted in theory by torchsparses' Sequential, SparseTensor and Conv3D.
Wanted to know any thoughts on that