EZ-SP: Fast and Lightweight Superpoint-Based 3D Segmentation

1 LIGM, ENPC, IP Paris, Univ Gustave Eiffel, CNRS, Marne-la-Vallée, France
2 DM3L, University of Zurich, Zurich, Switzerland

Summary

  • Superpoint-based methods are bottlenecked by the partition step.
  • EZ-SP: fast and learnable GPU-based partitioning.
  • 72× faster than point-based SOTA (PTv3) for end-to-end semantic segmentation.
  • 5.3× faster than superpoint-based SOTA (SPT) for end-to-end semantic segmentation.
  • Matching SOTA accuracy across domains (indoor, mobile mapping, aerial scanning).

Superpoint paradigm

3D semantic segmentation can be accelerated by grouping nearby points into semantically consistent regions, called superpoints. However, previous partitioning algorithms are slow due to being CPU-based.

First research result visualization
First research result visualization

Breaking the partition bottleneck

Breaking the partition bottleneck

Computation breakdown between SPT and EZ-SP on S3DIS 6-Fold.

First research result visualization

Inference speed v.s. performance v.s. model size of end-to-end pipelines (preprocessing to inference) on S3DIS. EZ-SP achieves near-SOTA accuracy with only 400k parameters, while being orders of magnitude faster than point-based networks.

Pipeline

EZ-SP Pipeline

EZ-SP. A 60k-parameter backbone embeds every point of the input scene into a low-dimensional space where adjacent points from different semantic classes are pushed apart. A GPU-accelerated algorithm then clusters neighbouring points with similar embeddings, producing semantically homogeneous superpoints. Finally, a lightweight (330k-parameter) superpoint-level network assigns a label to each superpoint, which is broadcast back to its points for dense segmentation.

Video Presentation

BibTeX

@article{geist2025ezsp,
  title={EZ-SP: Fast and Lightweight Superpoint-Based 3D Segmentation},
  author={Geist, Louis and Landrieu, Loic and Robert, Damien},
  journal={arXiv},
  year={2025},
}