📦 EVLab-SHPCD: A Cross-Modal Change Detection Dataset

July 27, 2025 · View on GitHub

EVLab-SHPCD is a cross-modal change detection benchmark proposed and led by Prof. Xiangyun Hu, and constructed by Dr. Kai Deng and other members of the Earth Vision Laboratory (EVLab) at Wuhan University, Hubei Province, China. The dataset is designed to promote research on detecting land use changes by combining historical land use base maps with current high-resolution remote sensing imagery.

🗂️ Dataset Overview

Training Set: 5,225 image pairs
Validation Set: 397 image pairs
Image Size: 512 × 512 pixels
Spatial Resolution: 0.8 meters

Each image pair consists of the following three components:

Hlubm (Historical Land Use Base Map)
- A grayscale image representing historical land use categories
- Pixel values range from 0 to 85, each corresponding to a specific land use type
Crism (Current Remote Sensing Image)
- A contemporary high-resolution optical remote sensing image
Target (Change Annotation)
- A binary label map manually annotated
- Pixel value 0: background / no change
- Pixel value 255: newly developed land, including buildings, roads, and filled areas

🧭 Supporting Files

trans.xml
- A raster code-to-class mapping file
- Defines the mapping between grayscale pixel values in Hlubm and semantic land use labels
dlbm.xml
- A category definition file following China's Third National Land Use Survey (第三次全国国土调查)
- Provides standardized land use labels for compatibility and extension

📥 Download

Baidu Cloud: https://pan.baidu.com/s/1Bq_P7wr5z6d3UixCyTtPtA
Access Code: cmcd

Note: This dataset is released strictly for academic research purposes only. Any commercial use is prohibited.

📖 Citation

If you use EVLab-SHPCD in your work, please cite the following paper:

@article{DENG2024114,
  title     = {Cross-modal change detection using historical land use maps and current remote sensing images},
  journal   = {ISPRS Journal of Photogrammetry and Remote Sensing},
  volume    = {218},
  pages     = {114-132},
  year      = {2024},
  issn      = {0924-2716},
  doi       = {https://doi.org/10.1016/j.isprsjprs.2024.10.010},
  url       = {https://www.sciencedirect.com/science/article/pii/S0924271624003873},
}