HouseNet (select preprocessed models from BuildingNet 0.1)

HouseNet contains select 'house' and 'villa' models that originated from the BuildingNet v0.1 data set (Pratheba, et al.) which were then preprocessed to remove site geometry and relabeled to only include the 'door', 'roof', 'wall', and 'window' labels since these labels were in all models. For each building in this data set, the following files are included:1) a modified point cloud file (.ply) which came from the original data set and was modified by removing site geometry and relabeling the points to make the models suited for deep learning for generative design2) a modified file containing label information (.json) which came from the original data set and was modified to make the models suited for deep learning for generative design3) a pickle file (.pkl) generated through pre-processing which contains information about the point cloud including the file name, label array, color array, typology, and building type information<br>Pratheba Selvaraju, Mohamed Nabail, Marios Loizou, Maria Maslioukova, Melinos Averkiou, Andreas Andreou, Siddhartha Chaudhuri, Evangelos Kalogerakis, "BuildingNet: Learning to Label 3D Buildings", <em>Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)</em>, 2021. https://buildingnet.org/HouseNet 数据集源自 BuildingNet v0.1 数据集（Pratheba 等），从中精选了符合要求的住宅（house）与别墅（villa）模型；由于该四类标签在所有模型中均覆盖，随后对所选模型进行预处理，移除场地几何信息，并仅保留门（door）、屋顶（roof）、墙体（wall）与窗户（window）四类语义标签。 本数据集为每个建筑提供以下三类文件： 1) 经修改的点云（point cloud）文件（.ply）：源自原始数据集，经移除场地几何信息、重标注点云语义标签后优化，适配生成式设计（generative design）相关深度学习任务； 2) 经修改的标签信息文件（.json）：源自原始数据集，经优化后适配生成式设计相关深度学习任务； 3) 预处理生成的 pickle 格式文件（.pkl）：包含点云相关信息，具体包括文件名、标签数组、颜色数组、建筑类型学（typology）与建筑类别信息。 该数据集的原始研究来自：Pratheba Selvaraju、Mohamed Nabail、Marios Loizou、Maria Maslioukova、Melinos Averkiou、Andreas Andreou、Siddhartha Chaudhuri、Evangelos Kalogerakis，论文《BuildingNet: Learning to Label 3D Buildings》，发表于2021年IEEE/CVF国际计算机视觉大会（ICCV）论文集，官方网站：https://buildingnet.org/