GabFormer

The official implementation of **Gabor Feature Network for Transformer-based Building Change Detection Model in Remote Sensing** (ICIP 2024)

Link of the paper

🍙 Dependencies

You can create an Anaconda environment named GabFormer from environment.yml.

conda env create -f environment.yml
conda activate GabFormer

🍴 Installation

Clone this repo:

git clone https://github.com/Ayana-Inria/GabFormer.git
cd GabFormer

🍞 Training

1. Change the dataset path in data_config.py.
2. Set the training parameter and model's hyperparameters in the training script run_GabFormer.sh in the folder scripts. You can run the script file by sh scripts/run_GabFormer.sh in the command environment.

🍨 Inference

You can find the evaluation script eval_GabFormer.sh in the folder scripts. You can run the script file by sh scripts/eval_GabFormer.sh in the command environment.

🍭 Dataset Preparation

🍵 Data structure

"""
data structure
-dataroot
    ├─A
        ├─img1.png
        ...
    ├─B
        ├─img1.png
        ...
    ├─label
        ├─img1.png
        ...
    └─list
        ├─val.txt
        ├─test.txt
        └─train.txt

# In list/ folder, prepare text files of the splits and list down all filenames of each split
   # for example:
       list/train.txt
           img1.png
           img32.png
           ...
       list/test.txt
           img2.png
           img15.png
           ...
       list/val.txt
           img54.png
           img100.png
           ...
"""

A: pre-change images;

B: post-change images;

label: binary labels;

list: contains train.txt, val.txt and test.txt, each file records the image names (XXX.png) in the change detection dataset.

🍪 Dataset

GabFormer was tested on two datasets for remote sensing building change detection.

1. LEVIR-CD

   • Paper: A Spatial-Temporal Attention-Based Method and a New Dataset for Remote Sensing Image Change Detection
   • Download: Link

2. WHU-CD

   • Paper: Fully convolutional networks for multisource building extraction from an open aerial and satellite imagery data set
   • Download: Link

We utilized the same cropping method as what is provided here to the size of 256 x 256.

License

The code is released under the GPL-3.0-only license. See LICENSE file for more details.

🌝 Citation

If you use this code for your research, please cite our paper (to be updated):

@inproceedings{GabFormer,
      title={Gabor Feature Network for Transformer-based Building Change Detection Model in Remote Sensing}, 
      author={Priscilla Indira Osa and Josiane Zerubia and Zoltan Kato},
      year={2024},
      
}

🐾 Acknowledgement

Our GabFormer is implemented based on the code provided in repositories below:

• https://github.com/wgcban/ChangeFormer
• https://github.com/jxgu1016/Gabor_CNN_PyTorch