Do 3D Point Representations Encode Architectural Functions? Probing Supervised and Pretrained Models on Urban Facades
Organizations: Graphics and Spatial Computing Lab, Saint Mary’s University, Halifax, Canada
Abstract
Semantic labels say what a building element is, not what it does: two identical-looking windows on one facade may be operable or fixed, a column may carry load or only decorate. We ask whether 3D point representations encode such architectural function. On 22 mobile-LiDAR buildings with 15 functional attributes per element, we hold out entire buildings and compare hand-crafted descriptors, networks trained from scratch and frozen self-supervised point transformers. Descriptors and frozen features are read by the same linear layer. No network trained from scratch, from PointNet to a 108M-parameter PTv3, clearly exceeds the descriptors (0.610 mAP on functionally ambiguous elements), whereas frozen Utonia reaches 0.732. Random-weight controls of the same architectures (0.649 and 0.652) attribute the gain to pretraining (Utonia: +0.080, 95% building interval [+0.035, +0.124]). Pretrained features improve at every tested number of labelled buildings while descriptors and random features plateau, and, in the configurations we tested, fine-tuning erodes the benefit: at today's data scale, function is better read from pretrained representations than learned from scratch.
Figures & tables
| Features | Network | Venue | Pretraining data | Trained | mAP | vs. random |
| Class only | one-hot | – | – | linear | 0.556 .011 | [ , ] |
| Hand-crafted | 21 descriptors | – | – | linear | 0.610 .004 | [ , ] |
| From scratch | PointNet [ 16 ] | CVPR’17 | – | all ( 1M) | 0.612 .018 | [ , ] |
| PointNet++ [ 17 ] | NeurIPS’17 | – | all ( 1M) | 0.622 .014 | [ , ] | |
| DGCNN [ 21 ] | TOG’19 | – | all ( 1M) | 0.613 .014 | [ , ] | |
| PTv3 [ 23 ] | CVPR’24 | – | all (108M) | 0.548 .012 | [ , ] |
Point Cloud Understanding
Point cloud understanding turns sparse 3D measurements into object, scene, shape, and motion information for machines.
See all 50 papers