/GAS-Net

A PyTorch implementation of GAS-Net based on the paper "Global-Aware Siamese Network for Change Detection on Remote Sensing Images"

Primary LanguagePython

GAS-Net

The pytorch implementation for Global-Aware Siamese Network for Change Detection on Remote Sensing Images on ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING.

The GAS-Net is designed to generate global-aware features for efficient change detection by incorporating the relationships between scenes and foregrounds.

Results

image1

image2

Requirements

  • Python 3.6
  • Pytorch 1.2.0

Datasets

The data folder is structured as follows:

├── data/
│   ├── levir_CD/    # Levir-CD dataset
|   |   ├── train/    # traning set 
|   |   |   ├── t1/    #images of time t1
|   |   |   ├── t2/    #images of time t2
|   |   |   ├── label/    #ground truth
|   |   ├── val/    # validation set
|   |   |   ├── t1/
|   |   |   ├── t2/
|   |   |   ├── label/
|   |   ├── test/    # testing set
|   |   |   ├── t1/
|   |   |   ├── t2/
|   |   |   ├── label/    #ground truth for evaluation
|   |   ├── results/    # path to save the model
│   ├── SVCD/
|   |   ├── leveb/    # Lebediv-CD dataset, have the same structure of the Levir-CD dataset
...

Citation

@article{Global2023zhang,
    title = {Global-aware siamese network for change detection on remote sensing images},
    journal = {ISPRS Journal of Photogrammetry and Remote Sensing},
    volume = {199},
    pages = {61-72},
    year = {2023},
    doi = {https://doi.org/10.1016/j.isprsjprs.2023.04.001},
    author = {Ruiqian Zhang and Hanchao Zhang and Xiaogang Ning and Xiao Huang and Jiaming Wang and Wei Cui},
}

Acknowledgment

This code is heavily borrowed from SRCDNet.