Block sparsity on input features

Thanks for the great effort!

I wonder if this library only supports block sparsity on weights? Or also the input features?

For example, can Sparse Convs operate over dense weight but (block) spatially sparse input feature maps?

