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

TL;DR

EZ-SP introduces a GPU-based superpoint partitioning method that is significantly faster, lightweight, and capable of real-time 3D segmentation, matching state-of-the-art accuracy across various 3D data domains.

Contribution

The paper presents a fully GPU learnable partitioning algorithm for superpoints, enabling fast, lightweight, and accurate 3D segmentation without handcrafted features.

Findings

13× faster superpoint partitioning than prior methods

Full pipeline fits in less than 2 MB VRAM and supports real-time inference

Achieves comparable accuracy to state-of-the-art point-based models across multiple datasets.

Abstract

Superpoint-based pipelines provide an efficient alternative to point- or voxel-based 3D semantic segmentation, but are often bottlenecked by their CPU-bound partition step. We propose a learnable, fully GPU partitioning algorithm that generates geometrically and semantically coherent superpoints 13$\times$ faster than prior methods. Our module is compact (under 60k parameters), trains in under 20 minutes with a differentiable surrogate loss, and requires no handcrafted features. Combine with a lightweight superpoint classifier, the full pipeline fits in $<$2 MB of VRAM, scales to multi-million-point scenes, and supports real-time inference. With 72$\times$ faster inference and 120$\times$ fewer parameters, EZ-SP matches the accuracy of point-based SOTA models across three domains: indoor scans (S3DIS), autonomous driving (KITTI-360), and aerial LiDAR (DALES). Code and pretrained models are accessible at github.com/drprojects/superpoint_transformer.