ZAHA
收藏资源简介:
ZAHA数据集是由慕尼黑工业大学创建的,是目前最大的3D立面语义分割数据集,包含601亿个标注点。数据集涵盖了多种建筑风格的66个立面,提供了15个与立面相关的类别。数据集的创建基于国际城市建模标准,确保了与现实世界挑战性类别的兼容性和方法的统一比较。数据集的创建过程利用了移动激光扫描(MLS)设备获取的密集街道级点云数据,并进行了详细的语义标注。ZAHA数据集主要应用于3D立面语义分割和城市数字孪生的创建,旨在解决立面语义分割中的挑战性问题。
The ZAHA Dataset, developed by the Technical University of Munich, is currently the largest 3D facade semantic segmentation dataset. It contains 60.1 billion annotated points, covering 66 facades across diverse architectural styles and providing 15 facade-related semantic categories. Built in compliance with international urban modeling standards, the dataset ensures compatibility with challenging real-world semantic categories and enables unified comparative evaluations of different methods. Its development utilizes dense street-level point cloud data acquired via Mobile Laser Scanning (MLS) equipment, accompanied by thorough semantic annotation. The ZAHA Dataset is primarily applied to 3D facade semantic segmentation and urban digital twin construction, with the goal of addressing challenging problems in facade semantic segmentation.
ZAHA 数据集概述
数据集简介
- 名称: ZAHA
- 类型: 点云数据集
- 用途: 立面语义分割
- 规模: 包含 601 百万个标注点
- 特点:
- 引入 LoFG(Level of Facade Generalization),支持立面的层次化理解
- 包含多种建筑风格
- 提供本地和全球(UTM)坐标参考系统
- 文件名指向巴伐利亚官方 CityGML LoD2 建筑模型
数据下载
- 下载链接: 下载地址
- 密码: zahahadid
数据集亮点
- 601 百万标注点
- 引入 LoFG:Level of Facade Generalization,支持立面的层次化理解
- 多种建筑风格
- 本地和全球(UTM)坐标参考系统
- 文件名指向官方 CityGML LoD2 建筑模型
立面语义分割结果
LoFG3 结果
| 模型 | OA | P | R | F1 | IoU |
|---|---|---|---|---|---|
| PointNet | 59.9 | 46.1 | 42.2 | 38.7 | 26.4 |
| PointNet++ | 66.4 | 37.8 | 35.9 | 34.8 | 25.6 |
| Point Transformer | 75.0 | 52.7 | 54.7 | 52.1 | 41.6 |
| DGCNN | 71.1 | 53.6 | 45.8 | 44.5 | 33.4 |
LoFG2 结果
| 模型 | OA | P | R | F1 | IoU |
|---|---|---|---|---|---|
| PointNet | 71.9 | 69.6 | 68.1 | 68.1 | 55.8 |
| PointNet++ | 75.5 | 73.0 | 73.0 | 72.6 | 59.8 |
| Point Transformer | 78.2 | 75.8 | 76.6 | 76.1 | 63.9 |
| DGCNN | 82.6 | 80.0 | 81.8 | 80.4 | 68.5 |
引用
plain @article{wysockietalZAHA, author = {Wysocki, O. and Tan, Y. and Froech, T. and Xia, Y. and Wysocki, M. and Hoegner, L. and Cremers, D. and Holst Ch.}, title = {ZAHA: Introducing the Level of Facade Generalization and the Large-Scale Point Cloud Facade Semantic Segmentation Benchmark Dataset}, booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)}, year = {2025}, }
plain @misc{wysocki2024zahaintroducinglevelfacade, title={ZAHA: Introducing the Level of Facade Generalization and the Large-Scale Point Cloud Facade Semantic Segmentation Benchmark Dataset}, author={Olaf Wysocki and Yue Tan and Thomas Froech and Yan Xia and Magdalena Wysocki and Ludwig Hoegner and Daniel Cremers and Christoph Holst}, year={2024}, eprint={2411.04865}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2411.04865}, }




