Multi scale method improves 3D point cloud segmentation without labels
MSSP: Multi-Scale Spatially-Constrained Partition for Unsupervised Semantic Segmentation of 3D Point Clouds
Computer Vision and Pattern RecognitionArtificial Intelligence
Summary
3D point clouds are collections of points that represent spaces or objects, and understanding their parts automatically is hard because labeling data takes a lot of work. The authors found that looking at these point clouds at multiple levels of detail helps capture complex structures, but only when combined with rules that keep nearby points grouped together. Their approach, called MSSP, uses multi-scale analysis with spatial constraints to group points more accurately without needing labels. Tests showed it performs better than previous automatic methods, especially in indoor scenes.
3D point cloudsemantic segmentationunsupervised learningmulti-scale analysisspectral clusteringsuperpointsspatial coherenceclusteringmIoUS3DIS
Authors
Zhenghao Zhang, Xinjie Wang, Wei Wang, Jun Zhang, Hanyun Wang
Abstract
3D point cloud semantic segmentation is essential for real-world spatial understanding, yet the prohibitive cost of human annotations motivates unsupervised approaches that require no labels. Existing superpoint-based methods typically rely on spectral analysis at a fixed granularity, failing to capture the hierarchical semantic structures inherent in complex indoor scenes. To bridge this gap, we present a Multi-Scale Spatially-Constrained Partition (MSSP) framework that combines multi-scale spectral analysis with spatially-constrained clustering. Multi-scale spectral analysis constructs enriched superpoint descriptors across multiple clustering granularities; however, the resulting high-dimensional feature space calls for a structural prior to translate into cleaner segmentation. Spatially-constrained clustering supplies this prior by restricting superpoint merging to physically adjacent regions, imposing the spatial coherence needed for multi-scale features to be effective. Extensive experiments on S3DIS and ScanNet show that MSSP achieves the best mIoU among unsupervised methods on the main benchmarks, with particularly significant gains on S3DIS. Notably, our ablation reveals a regularize-then-enrich interaction: multi-scale features alone do not improve final segmentation, yet become highly effective when coupled with spatial regularization, underscoring that spatial coherence is aprerequisite for multi-scale representations in superpoint clustering.