Text this: Efficient lower-limb segmentation for large-scale volumetric CT by using projection view and voxel group attention.