AI RESEARCH
Spira: Exploiting Voxel Data Structural Properties for Efficient Sparse Convolution in Point Cloud Networks
arXiv CS.LG
•
ArXi:2511.20834v2 Announce Type: replace-cross Sparse Convolution (SpC) powers 3D point cloud networks widely used in autonomous driving and augmented/virtual reality. SpC builds a kernel map that s mappings between input voxel coordinates, output coordinates, and weight offsets, then uses this map to compute feature vectors for output coordinates.