RandLA-Net: Efficient Semantic Segmentation of Large-Scale Point Clouds

By Qingyong Hu, Bo Yang, Linhai Xie, Stefano Rosa, Yulan Guo, Zhihua Wang, Agathoniki Trigoni, Andrew Markham. Computer Vision and Pattern Recognition, 2019

2,046 citations recorded by the source corpus.

We study the problem of efficient semantic segmentation for large-scale 3D point clouds. By relying on expensive sampling techniques or computationally heavy pre/post-processing steps, most existing approaches are only able to be trained and operate over small-scale point clouds. In this paper, we introduce RandLA-Net, an efficient and lightweight neural architecture to directly infer per-point semantics for large-scale point clouds. The key to our approach is to use random point sampling instead of more complex point selection approaches. Although remarkably computation and memory efficient, random sampling can discard key features by chance. To overcome this, we introduce a novel local feature aggregation module to progressively increase the receptive field for each 3D point, thereby effectively preserving geometric details. Extensive experiments show that our RandLA-Net can process 1 million points in a single pass with up to 200x faster than existing approaches. Moreover, our RandLA-Net clearly surpasses state-of-the-art approaches for semantic segmentation on two large-scale benchmarks Semantic3D and SemanticKITTI.

Subfields: Aerospace Engineering, Computational Mechanics, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/587110558b1744b912108dd0b74970f131317e0d

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Aerospace Engineering

In Computational Mechanics

In Computer Vision and Pattern Recognition

Citations in the corpus

Searching Efficient 3D Architectures with Sparse Point-Voxel Convolution (2020, neutral): “9 RandLA-Net [19] 1.”
Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework (2022, background): “…cloud analysis has seen a great improvement in various applications, including 3D shape classiﬁcation (Qi et al., 2017a), semantic segmentation (Hu et al., 2020) and object detection (Shi & Rajkumar, 2020), etc. Recent efforts have shown promising results for point cloud analysis by exploring…”
SUM: A Benchmark Dataset of Semantic Urban Meshes (2021, methodology): “To evaluate and compare with the current state-of-the-art 3D deep learning methods that can be applied to large-scale urban dataset, we select five representative approaches (PointNet [14], PointNet++ [46], SPG [15], KPConv [1], and RandLA-Net [47]), and we perform the experiment on an NVIDIA GEFORC”
Deep Learning for Image and Point Cloud Fusion in Autonomous Driving: A Review (2020, methodology): “To reduce the computational and memory cost of the PointNet++ [30], the RandLA-Net [39] stacked the random point sampling modules and attention-based local feature aggregation modules hierarchically to progressively increase receptive field while maintaining high efficiency.”
Voxel R-CNN: Towards High Performance Voxel-based 3D Object Detection (2020, methodology): “(Lang et al. 2019; Shi et al. 2020a; Yang et al. 2020; Ye, Xu, and Cao 2020).”
Campus3D: A Photogrammetry Point Cloud Benchmark for Hierarchical Understanding of Outdoor Scene (2020, background): “Categories of pointbased deep learning models mainly include feature pooling models [15, 23, 24, 43], convolution-based models [20, 30, 34], graph-based models [18, 33, 37] and attention-based models [41, 42].”
PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies (2022, neutral): “For example, KPConv [47] randomly drops colors during training, Point-BERT [60] uses a common point resampling strategy to randomly sample 1, 024 points from the original point cloud for data scaling, while RandLA-Net [15] and Point Transformer [62] load the entire scene as input on segmentation tas”
PolarNet: An Improved Grid Representation for Online LiDAR Point Clouds Semantic Segmentation (2020, neutral): “RandLA [12]: Hu et al.”
Sparse Single Sweep LiDAR Point Cloud Segmentation via Learning Contextual Shape Priors from Scene Completion (2020, methodology): “Among these methods, graph-based learning (Wang et al. 2019a; Landrieu and Si-monovsky 2018; Landrieu and Boussaha 2019; Wang et al. 2019c) and convolution-like operations (Thomas et al. 2019; Wu, Qi, and Fuxin 2019; Hu et al. 2020) are widely used.”
Walk in the Cloud: Learning Curves for Point Clouds Shape Analysis (2021, methodology): “RandLA-Net [6] reviews multiple sampling techniques and adopts the random sampling for efﬁciently processing large scale point clouds.”