VASAD: a Volume and Semantic dataset for Building Reconstruction from Point Clouds

VASAD: a Volume and Semantic dataset for Building Reconstruction from Point Clouds
P-A. Langlois, A. Boulch, and R. Marlet. VASAD: Volumetric and Semantic Reconstruction, ICPR, 2022.

Dataset download

You can download VASAD on Kaggle.

Dataset description

Each folder contains the data for a given building. We provide the following files:

• model.obj: the raw file where each component of the building is represented separately.
• colored_triangles.obj: each class of the dataset is colored for visualization purpose.
• ptView.json: a set of simulated camera locations
• samplesWithLabel.obj: a simulated Lidar scan (from the point of views in ptView.json) with a class label for each point.
• pov.obj: the list of point of views from which each point of samplesWithLabel.obj was taken from.

We also provide a labelled point cloud with features computed from FKAConv:

• lcpSamplesWithLabel.obj: a simulated Lidar scan (from the point of views in ptView.json with a FKAConv feature for each
• lcpPov.obj: the list of point of views from which each point of samplesWithLabel.obj was taken from.
• lcpVis.obj: a colored version of lcpSamplesWithLabel.obj using a argmax function on the features for visualization purpose.

The following classes are represented in the dataset:

• pillar
• railing
• roof
• stair
• window
• beam
• door
• partition
• slab
• bearing wall
• ceiling