issai/CSDS_dataset

If you use the dataset/source code/pre-trained models in your research, please cite our work. The preprint is available at this link.

🛰️ CSDS: AI-Based Construction Site Detection and Segmentation Tool for Satellite Images

The Construction Site Detection and Segmentation(CSDS) is a large-scale dataset of construction site satellite imagery with detailed polygon annotations.
It contains both the raw source data (images and XML annotations) and preprocessed training-ready splits in YOLO format.

---

Dataset Structure

Notes

• All images and annotations are provided in ZIP archives for efficient storage and download.
• The raw/ folder contains original images and XML annotations.
• The preprocessed/ folder contains processed input images (600px and 1200px) and corresponding annotaions in YOLO-style train/test/val splits.

Raw Data

raw/
├── images/                     # Original construction site images
└── annotations/
├── AOD/                     # XML annotations for *All Objects Dataset*
└── FVOD/                    # XML annotations for *Fully Visible Objects Dataset*

• AOD: includes all annotated objects, even partially occluded ones.
• FVOD: includes only fully visible objects.

Preprocessed Data

preprocessed/
├── AOD/
│   ├── 600/                     # Images with input resolution of 600px (YOLO format)
│   │   ├── train/
│   │   │   ├── images/
│   │   │   └── labels/
│   │   ├── val/
│   │   │   ├── images/
│   │   │   └── labels/
│   │   └── test/
│   │       ├── images/
│   │       └── labels/
│   └── 1200/                    # Images with input resolution of 1200px (YOLO format)
│       └── (train/test/val structure as above)
│
└── FVOD/
├── 600/
│   └── (train/test/val with images + labels)
└── 1200/
└── (train/test/val with images + labels)

• AOD/ → Preprocessed dataset corresponding to "all objects" annotations.
• FVOD/ → Preprocessed dataset corresponding to "fully visible objects" annotations.
• Each size folder (600/, 1200/) contains YOLO-ready train/, val/, and test/ splits with images/ and labels/ directories.

---

Intended Uses

• Training and evaluating computer vision models for object detection and segmentation in construction environments.
• Benchmarking performance on occlusion-aware vs fully-visible annotations.
• Studying dataset preprocessing effects (image resolution: 600px vs 1200px).

---

📄 Reference and Citation

This dataset accompanies the preprint:

Ulzhan Bissarinova, Hamad Hassan Awan, Sakiru Olarewaju Olagunju, et al. CSDS: AI-Based Construction Site Detection and Segmentation tool for Satellite Images. TechRxiv. October 15, 2025. DOI: 10.36227/techrxiv.176054630.07365932/v1

If you use this code or build upon our methods, please cite the preprint. If you use or analyze the dataset directly, please also cite the dataset DOI.

📊 Dataset

The dataset used for training and evaluation is registered at DOI:
👉 [https://doi.org/10.48333/0PJD-BP65]

BibTeX citation:

@article{Bissarinova_2025,
title={CSDS: AI-Based Construction Site Detection and Segmentation tool for Satellite Images},
url={http://dx.doi.org/10.36227/techrxiv.176054630.07365932/v1},
DOI={10.36227/techrxiv.176054630.07365932/v1},
publisher={Institute of Electrical and Electronics Engineers (IEEE)},
author={Bissarinova, Ulzhan and Awan, Hamad Hassan and Olagunju, Sakiru Olarewaju and Bolatkhanov, Iskander and Turekhassim, Abylay and Varol, Huseyin Atakan and Karaca, Ferhat},
year={2025},
month=oct }

---

Pretrained Models

Pretrained models trained on this dataset are available at:
👉 issai/CSDS_models

---

License

📜 MIT

---

Acknowledgements

This dataset was prepared by the Institute of Smart Systems and Artificial Intelligence (ISSAI) and Department of Civil and Environmental Engineering, Nazarbayev University.