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
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