Rohbau3D: A Shell Construction Site 3D Point Cloud Dataset

<p><strong>Rohbau3D</strong> is a dataset of real-world indoor construction environments represented by high-resolution 3D point clouds. It comprises 504 LiDAR scans captured with a terrestrial laser scanner across 14 distinct construction sites, including residential buildings, a large-scale office complex, educational facilities, and an underground parking garage, all in various stages of shell construction or renovation. Each point cloud is enriched with scalar laser reflection intensity, RGB color values, and reconstructed surface normal vectors. In addition to the native 3D data, the dataset includes high-resolution 2D panoramic renderings of each scene and its associated point cloud features.</p>

<p>The current release further extends Rohbau3D with dense point-wise semantic class labels and instance annotations for the point clouds, together with corresponding projected annotations in the panoramic renderings. The semantic taxonomy comprises 17 foreground classes and one background class, covering structurally relevant and frequently occurring object categories in shell construction environments. These annotations enable research on semantic segmentation, instance-level scene understanding, and object-wise analysis of real construction-site data.</p>

<p>Designed to reflect the complexity and variability of real-world construction sites, Rohbau3D captures clutter, occlusions, unfinished structures, temporary installations, and characteristic sensing artifacts of terrestrial laser scanning. Rohbau3D is intended to support research in geometric processing, 3D scene understanding, and intelligent computing in structural and civil engineering, while also providing a challenging benchmark for robust perception in complex built environments.</p>

<p>If you find our work useful in your research, please cite our paper:<br>
<a href="https://doi.org/10.1038/s41597-025-05827-7">https://doi.org/10.1038/s41597-025-05827-7</a></p>

<p><strong>Rohbau3D</strong> 是一个以高分辨率3D点云为表现形式的真实室内建筑环境数据集。该数据集包含通过地面激光扫描仪采集的504次激光雷达(LiDAR)扫描数据，覆盖14个不同的建筑工地，场景涵盖住宅建筑、大型办公楼、教育设施及地下停车场，且均处于外壳结构施工或翻新的不同阶段。每个点云均附带激光反射强度标量值、RGB颜色值以及重建得到的表面法向量。除原生3D数据外，数据集还提供了每个场景的高分辨率2D全景渲染图及其关联的点云特征。</p>
<p>本次发布的Rohbau3D进一步新增了针对点云的密集逐点语义类别标签与实例标注，以及全景渲染图中对应的投影标注。其语义分类体系包含17个前景类别与1个背景类别，覆盖外壳施工场景中与结构相关且高频出现的物体类别。此类标注可支撑针对真实建筑工地数据的语义分割、实例级场景理解以及面向物体的分析研究。</p>
<p>Rohbau3D旨在还原真实建筑工地的复杂性与多变性，收录了杂乱场景、遮挡情况、未完工结构、临时设施以及地面激光扫描特有的感知伪影。该数据集既可用于结构工程与土木工程领域的几何处理、3D场景理解及智能计算相关研究，也可为复杂建筑环境中的鲁棒感知任务提供具有挑战性的基准测试集。</p>
<p>若您的研究中用到了本数据集，请引用以下论文：<br>
<a href="https://doi.org/10.1038/s41597-025-05827-7">https://doi.org/10.1038/s41597-025-05827-7</a></p>

搜集汇总

背景与挑战

背景概述

Rohbau3D是一个包含504个LiDAR扫描的高分辨率3D点云数据集，覆盖14个不同的建筑工地，每个点云包含激光反射强度、RGB颜色值和表面法向量等信息。数据集还提供了点级的语义类别标签和实例标注，支持语义分割、场景理解和物体分析等研究。