Domain-generalizable point cloud instance segmentation of bridge components using class-balanced dynamic thresholding

doi:10.1016/j.autcon.2025.106631

Domain-generalizable point cloud instance segmentation of bridge components using class-balanced dynamic thresholding

2025-11-04

0

PRE

AI

DOI：10.1016/j.autcon.2025.106631

Original

Original request for help

Save

Abstract

En 中文

• UDA method for DL-based instance segmentation of bridge point clouds under class imbalance and distribution shifts. • Automated pipeline for generating synthetic bridge point clouds with accurate instance-level annotations. • Class-balanced dynamic thresholding enabling adaptive self-training based on confidence distributions. • Consistent accuracy gains across four real-world datasets without manual annotations.

Keywords:

Bridge point cloud data

Instance segmentation

Unsupervised domain adaptation

Self-training

Synthetic data

Deep learning

Domain-generalizable point cloud instance segmentation of bridge components using class-balanced dynamic thresholding

Abstract

Journal

Organization