InLUT3D: Challenging real indoor dataset for point cloud analysis

TL;DR

This paper introduces the InLUT3D point cloud dataset, a comprehensive resource designed to advance the field of scene understanding in indoor environments and proposes metrics and benchmarking guidelines essential for ensuring trustworthy and reproducible results in algorithm evaluation.

Abstract

In this paper, we introduce the InLUT3D point cloud dataset, a comprehensive resource designed to advance the field of scene understanding in indoor environments. The dataset covers diverse spaces within the W7 faculty buildings of Lodz University of Technology, characterised by high-resolution laser-based point clouds and manual labelling. Alongside the dataset, we propose metrics and benchmarking guidelines essential for ensuring trustworthy and reproducible results in algorithm evaluation. We anticipate that the introduction of the InLUT3D dataset and its associated benchmarks will catalyse future advancements in 3D scene understanding, facilitating methodological rigour and inspiring new approaches in the field.

Figures (6)

• Figure 1: A sample taxonomy of point clouds
• Figure 2: Distribution of points' categories in S3DIS dataset.
• Figure 3: Distribution of point categories in the Semantics3D dataset.
• Figure 4: Visibility of points for single-setup point clouds. Green points are visible to the laser beam, red points are eclipsed.
• Figure 5: Point cloud converted to spherical projection.