S3DIS数据集

S3DIS

Our dataset is composed of five large-scale indoor areas from three different buildings, each covering approximately 1900, 450, 1700, 870 and 1100 square meters (total of 6020 square meters). These areas show diverse properties in architectural style and appearance and include mainly office areas, educational and exhibition spaces, and conference rooms, personal offices, restrooms, open spaces, lobbies, stairways, and hallways are commonly found therein. One of the areas includes multiple floors, whereas the rest have one. The entire point clouds are automatically generated without any manual intervention using the Matterport [1] scanner (only 3D point clouds; no images used by our method). Parts of these areas can be seen in Fig. 5.

我们的数据集由三个不同建筑的五个大型室内区域组成，每个区域大约覆盖1900、450、1700、870和1100平方米(总共6020平方米)。这些区域在建筑风格和外观上展示了不同的属性，主要包括办公区域、教育和展览空间，以及会议室、个人办公室、卫生间、开放空间、大堂、楼梯和走廊。其中一个区域包括多个楼层，而其他区域只有一层。使用Matterport扫描仪(仅使用3D点云;我们的方法没有使用图像)。这些区域的部分如图5所示。

We evaluated our method on a new dataset of several buildings with a covered area of over 6, 000m² and over 215 million points, demonstrating robust results readily useful for practical applications.

我们在一个新的数据集上对我们的方法进行了评估，该数据集覆盖面积超过6000平方米，超过2.15亿个点，展示了对实际应用非常有用的健壮结果。

We detect 12 semantic elements, which are structural elements (ceiling, floor, wall, beam, column, window and door) and commonly found items and furniture (table, chair, sofa, bookcase and board). Notice that these classes are more fine-grained and challenging than many of the semantic indoor segmentation datasets .

我们检测了12个语义元素，它们是结构元素(天花板、地板、墙壁、梁、柱、窗和门)和常见的物品和家具(桌子、椅子、沙发、书架和木板)。注意，这些类比许多语义室内分割数据集更细粒度和更具挑战性。

• 下载：https://shapenet.cs.stanford.edu/media/indoor3d_sem_seg_hdf5_data.zip

For scene segmentation, we first experiment with the S3DIS dataset [1], which has 13 categories of indoor scene objects. Each point has 9 attributes: XYZ coordinates, RGB color, and normalized coordinates w.r.t. the room space it belongs to.

对于场景分割，我们首先使用S3DIS数据集进行实验，该数据集包含13类室内场景对象。每个点有9个属性:XYZ坐标、RGB颜色和关于它所属的房间空间的归一化坐标。

• 数据维度：[XYZ, RGB, Normalized coordinates] (9个维度)

基于数据的分析

• h5文件(共24个，0~23)，[data, label]

  1553843064049.png

• data : (1000, 4096, 9)
• label : (1000, 4096)
• 第24个文件（23.h5）不足1000

第一个维度的含义

• 6个大场景中一个小房间的一部分
• 一个房间在第一维占x，则其点云数为x*4096
• 1553843478287.png

• 全部点云 ：23585 * 4096 = 96604160
• 房间数 ：272
• 平均每个房间点云数：355162