☀️S1GFloods Benchmark☀️

DAM-Net: Flood detection from SAR imagery using differential attention metric-based vision transformers

Tamer Saleh^1,2, Xingxing Weng¹, Shimaa Holail¹, Chen Hao¹, Gui Song-Xia¹

¹ Wuhan University, ² Benha University

🛎️Updates

• 🎉 February 2026 Achievement: S1GFloods has been selected as an 🔥 ESI Hot Paper and Highly Cited Paper, placing it among the top 1% of publications in the Geosciences field 🏆🌍
• 13 May 2024: DAM-Net has been accepted by ISPRS JP&RS and online available now!!
• 02 July 2023: The S1GFloods benchmark related to our paper has now released. You are warmly welcome to use it!!
• 25 June 2023: DAM-Net has been submitted for publication at ISPRS Journal of Photogrammetry and Remote Sensing!!
• 01 Jun 2023: The arXiv paper of DAM-Net is now online.

🔭Dataset Overview

• S1GFloods is the first open-access, globally distributed, event-diverse Sentinel-1 SAR dataset specifically designed to support AI-based flood response applications. The dataset comprises 5,360 image pairs with a spatial size of 256 × 256 pixels, covering 46 major flood events that occurred between 2015 and 2022 across six continents, with particular emphasis on developing countries. Its broad geographic and event diversity provides a comprehensive benchmark for developing and evaluating robust flood-mapping models.

• The accompanying animation illustrates pre- and post-event SAR imagery along with sample flood-mapping results for a rural region in Iran affected by severe flooding in March 2019. The figure on the right presents a magnified visualization of a 1 km × 1 km area (highlighted by the yellow box in the larger scene), demonstrating the extent of flood impacts on buildings as identified by our model.

• Dataset Statistics

• Training Set: 4,300 image pairs
• Validation Set: 530 image pairs
• Testing Set: 530 image pairs
• Image Size: 256 × 256 pixels
• Spatial Resolution: 10 meters

• Each image pair includes:

• A manually annotated binary label map
• Pixel value 0: Background / non-flooded area
• Pixel value 255: Newly inundated flood area

Requirements

Our model

An overview of the proposed DAM-Net. The feature maps of the pre-and post-event image pairs are extracted through a Siamese structure and pre-trained remote sensing.

🔭 Baselines

• 📖 📖 📖 DTCDSCN [here]
• 📖 📖 📖 UNet [here]
• 📖 📖 📖 FC-Siam [here]
• 📖 📖 📖 SNUNet–ECAM [here]
• 📖 📖 📖 Siam-Nested-UNet [here]
• 📖 📖 📖 ResNet50-IMP [here]
• 📖 📖 📖 ResNet50-RSP [here]
• 📖 📖 📖 Swin–T-RSP [here]
• 📖 📖 📖 Swin-T-IMP [here]
• 📖 📖 📖 ViTAEv2 [here]

📒 Dataset Preparation

Prepare the following folders to organize this repo:

 For the S1GFloods dataset, clip the images to 256 × 256 patches. Please, respect the following structure: 
          ├── SIGFloods
          │   ├── train
          │   │   ├── A                           Images of Time 1 before the flood event
          │   │   │   └── <region><year><XY>.png
          │   │   ├── B                           Images of Time 2 after the flood event
          │   │   │   └── <region><year><XY>.png
          │   │   └── GT                          Ground truth labels
          │   │       └── <region><year><XY>.png
          │   ├── val 
          │   │   ├── A                           
          │   │   │   └── <region><year><XY>.png
          │   │   ├── B                           
          │   │   │   └── <region><year><XY>.png
          │   │   └── GT                          
          │   │       └── <region><year><XY>.png
          │   ├── test
          │   │   ├── A                           
          │   │   │   └── <region><year><XY>.png
          │   │   ├── B                           
          │   │   │   └── <region><year><XY>.png
          │   │   └── GT                          
          │   │       └── <region><year><XY>.png
          │  

🚚 Datasets

You can download our novel public S1GFloods dataset through the following link:

• [S1GFloods]baidu drive
• [S1GFloods]Google Drive Link
• [S1GFloods]HuggingFace Link

📚 Use example

• Training

  python train.py

• Testing

  python eval.py

Results

Visualization

📃 Citing

@article{saleh2024dam,
  title={DAM-Net: Flood detection from SAR imagery using differential attention metric-based vision transformers},
  author={Saleh, Tamer and Weng, Xingxing and Holail, Shimaa and Hao, Chen and Xia, Gui-Song},
  journal={ISPRS Journal of Photogrammetry and Remote Sensing},
  volume={212},
  pages={440--453},
  year={2024},
  publisher={Elsevier}
}

Contact Information

If you have any questions or would like to collaborate, please reach out to me at tamersaleh@whu.edu.cn.

License

The datasets are released for non-commercial and research purposes only. For commercial purposes, please contact the authors.

Acknowledgment

Appreciate the work from the following repositories:

• wenhwu/awesome-remote-sensing-change-detection
• SNUNet-CD
• BIT-CD

Related resources

• ASF-Dataset
• Sentinel-Hub
• SNAP Toolbox

Star History