Semantic segmentation in construction environments—SegFormer checkpoint and synthetic training data

RWTH Publications (RWTH Aachen)

Abstract

This dataset contains synthetic training data for training neural networks for semantic segmentation in construction site environments. The data provides color images and semantic labels that were generated in simulation using domain randomization in the VEROSIM simulation platform. Additionally, a checkpoint of a SegFormer-B5 model partially trained on this data is included. This dataset was created as part of the “Off-Highway-Twins 2” research project and focuses on applications in road and underground utilities construction

Authors 4

1. Jorge Luis Jiménez-Aparicio Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen

2. Artem Bliznyuk Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen

3. Böken David Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen

4. Heinz-Jürgen Roßmann Aachen

   RWTH Aachen University

   Affiliation as printed

   RWTH Aachen