---
license: cc-by-4.0
language:
- en
tags:
- super-resolution
- hyperspectral
- remote_sensing
size_categories:
- 1K<n<10K
---
# GEWDiff Training & Evaluation Dataset

## 📘 Overview

The **GEWDiff Training & Evaluation Dataset** is derived from the EnMAP Champion and MDAS hyperspectral datasets.  
It is designed for image enhancement, super-resolution, restoration, and generative remote sensing tasks.  
The dataset includes Low-Quality (LQ) low-resolution images, corresponding Ground-Truth (GT) high-resolution images, and optional structure information such as **masks** and **edges** (partially provided; remaining components can be automatically generated using the accompanying GitHub scripts).

All data have been preprocessed, spatially tiled, spectrally unified, and harmonized through **nearest-neighbor approximation of the spectral response functions (SRF)**.

---

## 📂 Dataset Structure

### **1. Training Set**
- **LQ images**: low-quality / low-resolution observations  
- **GT images**: high-quality ground-truth targets  
- **Mask (partial)**: missing parts can be generated with included scripts  
- **Edge (partial)**: missing parts can be generated with included scripts  
Used for model training across various reconstruction and generative tasks.

---

### **2. Validation Set (val)**
- Same structure as the training set  
- Paired LQ–GT samples for model validation and tuning

---

### **3. Test Sets (with ground truth)**
Includes several subsets:

- **MDAS1**  
- **MDAS2**  
- **WDC**

These subsets contain paired LQ–GT data and are suitable for quantitative evaluations.

---



## 📐 Preprocessing Details

The dataset originates from **EnMAP Champion** and **MDAS hyperspectral** sources.  
All data have undergone:

- Spatial tiling  
- Spectral band unification  
- **Spectral response harmonization using nearest-neighbor approximation**  
- Conversion into LQ/GT pairs suitable for super-resolution, enhancement, and generative modeling tasks

---

## 🔧 Additional Resources

Mask and edge maps—when not provided—can be generated automatically using the scripts available in the linked GitHub repository.  
These structural cues enable models to leverage both texture and geometric information.

---

## 📑 Citation

If you use this dataset in your research or applications, please cite **our paper** (arXiv](https://arxiv.org/abs/2511.07103)):

```bibtex
@misc{wang2025gewdiffgeometricenhancedwaveletbased,
      title={GEWDiff: Geometric Enhanced Wavelet-based Diffusion Model for Hyperspectral Image Super-resolution}, 
      author={Sirui Wang and Jiang He and Natàlia Blasco Andreo and Xiao Xiang Zhu},
      year={2025},
      eprint={2511.07103},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/2511.07103}, 
}