Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

SemanticBridge -- A Dataset for 3D Semantic Segmentation of Bridges and Domain Gap Analysis

Maximilian Kellner, Mariana Ferrandon Cervantes, Yuandong Pan, Ruodan Lu, Ioannis Brilakis, Alexander Reiterer

TL;DR

The paper introduces SemanticBridge, a large-scale dataset for 3D semantic segmentation of bridges and analyzes how different 3D sensors (TLS vs MLS) induce domain gaps that affect model performance.It provides 20 bridges from the UK and Germany with 9 annotated classes, plus a practical interpolation method to extend TLS annotations to MLS data, enabling cross-sensor evaluation.Three state-of-the-art architectures (UNet3D, KPConv, PTv2) are benchmarked, revealing robust within-sensor performance but notable performance drops (6.9–11.4 mIoU points) when transferring to data from a different sensor, with varying sensitivity across classes.The work demonstrates the dataset’s potential for advancing domain adaptation and digital-twin research in infrastructure monitoring and sets the stage for broader sensor diversity in future studies.

Abstract

We propose a novel dataset that has been specifically designed for 3D semantic segmentation of bridges and the domain gap analysis caused by varying sensors. This addresses a critical need in the field of infrastructure inspection and maintenance, which is essential for modern society. The dataset comprises high-resolution 3D scans of a diverse range of bridge structures from various countries, with detailed semantic labels provided for each. Our initial objective is to facilitate accurate and automated segmentation of bridge components, thereby advancing the structural health monitoring practice. To evaluate the effectiveness of existing 3D deep learning models on this novel dataset, we conduct a comprehensive analysis of three distinct state-of-the-art architectures. Furthermore, we present data acquired through diverse sensors to quantify the domain gap resulting from sensor variations. Our findings indicate that all architectures demonstrate robust performance on the specified task. However, the domain gap can potentially lead to a decline in the performance of up to 11.4% mIoU.

SemanticBridge -- A Dataset for 3D Semantic Segmentation of Bridges and Domain Gap Analysis

TL;DR

The paper introduces SemanticBridge, a large-scale dataset for 3D semantic segmentation of bridges and analyzes how different 3D sensors (TLS vs MLS) induce domain gaps that affect model performance.It provides 20 bridges from the UK and Germany with 9 annotated classes, plus a practical interpolation method to extend TLS annotations to MLS data, enabling cross-sensor evaluation.Three state-of-the-art architectures (UNet3D, KPConv, PTv2) are benchmarked, revealing robust within-sensor performance but notable performance drops (6.9–11.4 mIoU points) when transferring to data from a different sensor, with varying sensitivity across classes.The work demonstrates the dataset’s potential for advancing domain adaptation and digital-twin research in infrastructure monitoring and sets the stage for broader sensor diversity in future studies.

Abstract

We propose a novel dataset that has been specifically designed for 3D semantic segmentation of bridges and the domain gap analysis caused by varying sensors. This addresses a critical need in the field of infrastructure inspection and maintenance, which is essential for modern society. The dataset comprises high-resolution 3D scans of a diverse range of bridge structures from various countries, with detailed semantic labels provided for each. Our initial objective is to facilitate accurate and automated segmentation of bridge components, thereby advancing the structural health monitoring practice. To evaluate the effectiveness of existing 3D deep learning models on this novel dataset, we conduct a comprehensive analysis of three distinct state-of-the-art architectures. Furthermore, we present data acquired through diverse sensors to quantify the domain gap resulting from sensor variations. Our findings indicate that all architectures demonstrate robust performance on the specified task. However, the domain gap can potentially lead to a decline in the performance of up to 11.4% mIoU.

Paper Structure

This paper contains 10 sections, 1 equation, 4 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Data Capturing
  4. Data Processing
  5. Experiments
  6. Architecture Details
  7. Implementation Details
  8. Comparison and Results
  9. Discussion
  10. Conclusion

Figures (4)

  • Figure 1: Visualization of point clouds showing the same scene captured by two different scanners. Zoom in for a better view.
  • Figure 2: Visualization of point clouds showing a colored bridge and a bridge which was annotated manually. The last shows an example of automatic label interpolation.
  • Figure 3: Visualization of point clouds showing a colored bridge and the randomly calculated center points for subcloud calculation.
  • Figure 4: Visualization of example predictions. The initial column depicts the outcomes obtained with the UNet3D, the subsequent column illustrates the results achieved with KPConv, and the final column presents the outcomes attained with PTv2. The top row of point clouds was captured using the stationary sensor, while the lower row was captured by the mobile scanner. Zoom in for a better view.