---
license: creativeml-openrail-m
task_categories:
  - image-to-image
  - image-segmentation
  - mask-generation
language:
  - en
tags:
  - satellite
  - remote-sensing
  - vhr
  - diffusion
  - inpainting
  - geospatial
dataset_name: KAO-DIFFSAT-VHR-BENCHMARK
---

# 🛰️ KAO-DIFFSAT-VHR-BENCHMARK  
**High-Resolution Satellite Image Dataset for Diffusion-Based Inpainting Research**

🚀 Project Page: https://kaopanboonyuen.github.io/KAO/

---

## 🌍 Overview

The **KAO-DIFFSAT-VHR-BENCHMARK** dataset is a curated collection of **very high-resolution (VHR) satellite imagery samples** designed for evaluating diffusion-based image inpainting methods in remote sensing.

This dataset is used in conjunction with our research:

> **KAO: Kernel-Adaptive Optimization in Diffusion for Satellite Image Inpainting**

📄 Paper: https://arxiv.org/abs/2511.02462  
🎉 Accepted to IEEE Transactions on Geoscience and Remote Sensing (TGRS)  
🌏 Presented at AOGS 2026, Japan

---

## 🛰️ Dataset Description

This dataset contains high-resolution satellite image samples used for:

- 🧠 Image inpainting benchmarking
- 🌍 Geospatial reconstruction tasks
- 🏙️ Urban / rural structure restoration
- 🌱 Land-cover consistency evaluation

---

## 📁 Dataset Structure

```

SAMPLE_SATELLITE_IMAGE/
SAMPLE_SATELLITE_IMAGE_0001.jpg
SAMPLE_SATELLITE_IMAGE_0002.jpg
SAMPLE_SATELLITE_IMAGE_0003.jpg
...

````

---

## 📊 Notes

- This dataset is intended for **research and benchmarking purposes**
- Images are used as **input samples for diffusion-based inpainting evaluation**
- Masks and ground-truth annotations may be released in future updates

---

## 📚 Citation

If you use this dataset, please cite:

```bibtex
@article{panboonyuen2025kao,
  title={KAO: Kernel-Adaptive Optimization in Diffusion for Satellite Image},
  author={Panboonyuen, Teerapong},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  year={2025},
  publisher={IEEE}
}
````

---

## 🙏 Acknowledgements

Parts of this dataset are inspired by:

```bibtex
@article{boguszewski2020landcoverai,
  title={LandCover.ai: Dataset for Automatic Mapping of Buildings, Woodlands, Water and Roads from Aerial Imagery},
  author={Boguszewski, Adrian and others},
  journal={arXiv preprint arXiv:2005.02264},
  year={2020}
}
```

We thank the authors of **LandCover.ai** for their valuable contribution to the remote sensing community.

---

## ⚠️ Disclaimer

This dataset is provided for academic and research use only.

---