TUM-FAÇADE

The latest, third version of the dataset, merges 'training' and 'test' datasets into one 'pointcloud' folder, to enable more flexibility. Previously non-annotated buildings (ID...57), are now labelled. Thus, our dataset is enriched with another 16 annotated facades, which translates to extra 77 mln annotated points. The missing class in the building ID...62 has been added, too. The updated statistics presents as follows: 33 annotated facades and 8 non-annotated for further benchmark extension or testing; 333 ml annotated points; in local and global coordinate reference system; settings file for adding your own data

本数据集的最新第三版将训练集与测试集合并为一个pointcloud（点云）文件夹，以提升使用灵活性。此前未完成标注的建筑（ID……57）现已完成全标注，由此新增16个标注立面，对应额外7700万标注点。建筑ID……62中缺失的类别也已补充完备。更新后的统计信息如下：共计33个标注立面与8个未标注立面，可用于后续基准测试拓展或测试；总标注点数量达3.33亿；支持局部与全局坐标参考系；附带可用于添加自定义数据的配置文件。

搜集汇总

背景与挑战

背景概述

TUM-FAÇADE是一个包含33个带注释立面图和3.33亿注释点的研究数据集，适用于语义分割和3D重建等应用。数据集基于TUM-MLS 2016点云数据，提供本地和全球坐标参考系统，支持进一步的研究和测试。