Images&PointClouds Cultural Heritage Dataset

The dataset comprises five heritage buildings, providing two types of data for each scene: photogrammetric images with corresponding intrinsic and extrinsic parameters, and the related point cloud. Annotated ground truth is available for both data types, ensuring high-quality training and validation resources. This dataset is particularly designed to support the training, validation, and testing of machine learning models. Its primary focus is to facilitate the automation of three-dimensional and informative model generation through semantic segmentation techniques. The multi-source nature of the dataset makes it highly versatile. It can be used to implement point-based and multi-view-based approaches, compare their performance on the same data source, and develop innovative hybrid networks that integrate both images and point clouds. The dataset adheres to the classification standards of ARCHdataset, defining the following classes: ‘arch’ (0), ‘column’ (1), ‘moulding’ (2), ‘floor’ (3), ‘door/window’ (4), ‘wall’ (5), ‘stair’ (6), ‘vault’ (7), ‘roof’ (8), and ‘other’ (9).

For more detailed information and to cite this dataset, please refer to:

Pellis, E., Masiero, A., Betti, M., Tucci, G., Grussenmeyer G. (2025). A photogrammetric image-point dataset for the semantic segmentation of heritage building. Data in Brief, Vol. 60. https:// doi.org/10.1016/j.dib.2025.111661.