Ziel unseres Berichts ist die Evaluation der Datenverfügbarkeit und das Erstellen eines Datensatzes, der später zum maschinellen Lernen von Bibliotheksklassifikationen genutzt werden könnte. Als Basis für die Textdaten werden wir Wikidata-Einträge nutzen, da diese teilweise bereits mit solchen Klassifikationen versehen und direkt mit dem zugehörigen Wikipedia-Artikel verknüpft sind.

Common Machine-Learning (ML) approaches for scene classification require a large amountof training data. However, for classification of depth sensor data, in contrast to image data, relativelyfew databases are publicly available and manual generation of semantically labeled 3D point clouds isan even more time-consuming task. To simplify the training data generation process for a wide rangeof domains, we have developed theBLAINDERadd-on package for the open-source 3D modelingsoftware Blender, which enables a largely automated generation of semantically annotated point-cloud data in virtual 3D environments. In this paper, we focus on classical depth-sensing techniquesLight Detection and Ranging (LiDAR) and Sound Navigation and Ranging (Sonar). Within theBLAINDERadd-on, different depth sensors can be loaded from presets, customized sensors can beimplemented and different environmental conditions (e.g., influence of rain, dust) can be simulated.The semantically labeled data can be exported to various 2D and 3D formats and are thus optimizedfor different ML applications and visualizations. In addition, semantically labeled images can beexported using the rendering functionalities of Blender.