3DCoMPaT++ extends large-scale 3D compositional recognition resources for studying materials, parts, and object-level visual understanding.
This work investigates scaling laws for PointNets and studies how improved training and scaling strategies affect 3D point-cloud understanding.
3D CoMPaT is a large-scale 3D vision dataset for studying the composition of materials on object parts. It supports compositional recognition over object geometry, parts, and material appearance, enabling research on richer 3D object understanding …
PointNet++ is one of the most influential neural architectures for point cloud understanding. Although its accuracy has been surpassed by recent networks such as PointMLP and Point Transformer, we find that much of the performance gain comes from …