Drop-in replacement PyTorch layer for sparse convolution. Calls an autograd-extended sparse convolution function that uses a CUDA backend. Backwards method is not supported currently.
from spconv.layer import SpConv2d
layer = SpConv2d.from_dense(my_conv2d_layer)
out = layer(in)
A more detailed sample that uses the functional interface can be found in correct_test.py
.