ConRebSeg: A Segmentation Dataset for Reinforced Concrete Construction

DTU Electro, Automation & Control Group

Patrick Schmidt and Lazaros Nalpantidis

---

This repository contains code to initizliaze the ConRebSeg dataset. The main component of the repository is the Python script import_dataset.py, which initializes the FiftyOne dataset and downloads the data for the dataset in a fully automated manner.

The accompanying self-collected data is indexed in the Technical University of Denmark (DTU) Data repository.

DOI: 10.11583/DTU.26213762

NB: Note that the initialization script will download from YouTube. Always ensure you comply with local laws regarding this process.

Initialization

To initialize the FiftyOne dataset and to download the data, please follow these steps:

1. Create a Python 3.12 virtual environment with Conda or venv and install the pip dependencies:

   pip install -r requirements.txt

2. Execute the initialization script. Note that this will download the self-collected data from DTU Data (~22 GB) and will unpack it in this directory under data/langebro and data/vester_sogade. It will also download videos from YouTube, extract frames and store them under data/youtube.

   python import_dataset.py

   The command above in general doesn't need to be modified. You can opt for the following options:

   • --skip_integrity_check: The script checks the integrity of each sample in the dataset to ensure that the download and import was execute without errors and that the data hasn't changed. It is not recommended to use this option, but it can save you time if needed.
   • --skip_yt_download: This disables the download process for YouTube videos. Useful if you're not certain about the legal status of downloading YouTube data in your jurisdiction.
   • --skip_selfcollected This disables the download of the self-collected sequences.

3. After the script has been executed successfully, open the FiftyOne app as follows:

   fiftyone app launch ConRebSeg

   This will open a browser window and present you with the dataset explorer. Happy exploring!

Further information

Further information about this dataset, its structure and characteristics can be found in the accompanying journal article Segmentation dataset for reinforced concrete construction - ScienceDirect

Contributors

Thanks to Rasmus E. Andersen, Javier Casas Lorenzo and Carlos Gascon Bononad for helping me in the collection process! Thanks to Christiansen & Essenbæk A/S for organizing access to the construction sites.

Citation

If you found this dataset and the article/paper useful, please cite us as follows:

@article{SCHMIDT2025105990,
title = {Segmentation dataset for reinforced concrete construction},
journal = {Automation in Construction},
volume = {171},
pages = {105990},
year = {2025},
issn = {0926-5805},
doi = {https://doi.org/10.1016/j.autcon.2025.105990},
url = {https://www.sciencedirect.com/science/article/pii/S0926580525000305},
author = {Patrick Schmidt and Lazaros Nalpantidis},
keywords = {Dataset, Construction robotics, Segmentation, Rebar detection, Shotcrete, Digitization}
}

Acknowledgements

This work has been funded and supported by the EU Horizon Europe project “RobetArme” under the Grant Agreement 101058731.